<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by David Okeyode on Medium]]></title>
        <description><![CDATA[Stories by David Okeyode on Medium]]></description>
        <link>https://medium.com/@davidokeyode?source=rss-77eea12d7ea0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*1eVrZnKX7Cjs5BqO8X35tA.png</url>
            <title>Stories by David Okeyode on Medium</title>
            <link>https://medium.com/@davidokeyode?source=rss-77eea12d7ea0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 09:50:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@davidokeyode/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AI Hands-On Lab for Security Engineers (3): Enterprise MCP Registry and Governance with Azure API…]]></title>
            <link>https://davidokeyode.medium.com/ai-hands-on-lab-for-security-engineers-3-enterprise-mcp-registry-and-governance-with-azure-api-f632b7e9c1fd?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/f632b7e9c1fd</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Mon, 30 Mar 2026 09:04:33 GMT</pubDate>
            <atom:updated>2026-03-30T09:04:33.784Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI Hands-On Lab for Security Engineers (3): Enterprise MCP Registry and Governance with Azure API Center</h3><p>Estimated Duration: 30–45 minutes | Difficulty: Intermediate</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IxiSx_9E9bL80mTkaqZaEQ.png" /></figure><h3>Background</h3><p>In <a href="https://davidokeyode.medium.com/ai-hands-on-lab-for-the-security-engineer-2-governing-mcp-servers-with-an-ai-gateway-in-azure-0101bd7376b3">Lab 2</a> you deployed a REST API (httpbin) to Azure App Service, imported it into Azure API Management, and exposed it as a governed MCP server with JWT authentication and rate limiting.</p><p>In this lab, we will address the challenge of governing MCP server consumption in the organization using a private MCP registry. This way, developers in the org can discover which MCP servers are available and approved, request for MCP servers to be validated and added. Without a central catalog, teams will independently adopt random public MCP servers, creating “shadow AI” (where everyone uses unreviewed and unsactioned tools that pose security and compliance risks.</p><p>We wil be using <strong>Azure API Center</strong> to address this use case. It provides a <strong>centralized inventory and discovery portal</strong> for all your MCP servers . You can think of it as your organization’s private MCP “app store” where developers browse, evaluate, and connect only to approved MCP servers.</p><h4>What You Will Learn and Implement</h4><ul><li>How Azure API Center serves as a private, enterprise-ready MCP registry</li><li>How to sync your APIM-managed MCP servers (including the httpbin server from Lab 2) into API Center automatically</li><li>How to use the built-in MCP discovery page to register supported partner MCP servers (e.g., Atlassian, Playwright)</li><li>How to deploy the API Center Portal for enterprise-wide MCP server discovery</li><li>How to connect your private registry to VS Code and GitHub Copilot</li></ul><h4>Prerequisites</h4><ul><li>Completed <a href="https://davidokeyode.medium.com/ai-hands-on-lab-for-the-security-engineer-2-governing-mcp-servers-with-an-ai-gateway-in-azure-0101bd7376b3">Lab 2</a> . You should have an APIM instance with the HTTPBin MCP server.</li><li>An Azure subscription with Contributor access.</li><li>A modern web browser.</li><li>VS Code with GitHub Copilot for the final integration exercise.</li></ul><p><strong>Note: </strong>The Free tier of Azure API Center is sufficient for this lab.</p><h4>1 — Create an Azure API Center Resource</h4><p>1. In the Azure Portal, search for <strong>API Centers</strong> and click <strong>+ Create</strong>.</p><p>2. Configure the following:</p><ul><li><strong>Resource group</strong>: MCPLabResources</li><li><strong>Name</strong>: Choose a unique name (e.g., mcp-registry-XXXX or your organization name)</li><li><strong>Region</strong>: Select your preferred region (e.g., East US)</li><li><strong>SKU</strong>: Free is sufficient for this lab (Standard for production)</li></ul><p>3. Click <strong>Review + create</strong> and then <strong>Create</strong>.</p><p>4. Wait for deployment to complete, then click <strong>Go to resource</strong>.</p><h4>2 — Sync MCP Servers from Azure API Management</h4><p>Now you will link your APIM instance to API Center so that your HTTPBin MCP server (and any other APIs) are automatically discovered and synchronized.</p><p>1. In your API Center resource, in the left menu select <strong>Platforms</strong> and then click <strong>Integrations</strong>.</p><p>2. Click <strong>+ New integration</strong> and select <strong>From Azure API Management</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vHK1RlfVNOmg-sTkdxrDJA.png" /></figure><p>3. Configure the integration form as follows:</p><ul><li><strong>Azure API Management service</strong>: Select the instance from Lab 2 in the dropdown</li><li><strong>Link identification</strong>: mcplab-api-XXXXXXXXX</li><li><strong>Environment title</strong>: Production</li><li><strong>Identification</strong>: production</li><li><strong>Environment type</strong>: Production</li><li><strong>Description</strong>: APIM Production environment</li><li><strong>Lifecycle</strong>: Production</li></ul><p>4. Leave other options as default and click <strong>Create</strong>.</p><p><strong>Note: </strong>The integration takes a minute or more to connect to the API Management instance, retrieve existing artifacts, and begin synchronization.</p><p>5. After a few moments, refresh the status. When it’s done you should see the state set to <strong>Linked and syncing</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wy0FRs55W0lW052bQpC-Ag.png" /></figure><h4>Verify the Sync</h4><p>1. In the left menu, under <strong>Inventory</strong>, select <strong>Assets</strong>.</p><p>2. You should see all the APIs from your APIM instance, including the <strong>HTTPBin API</strong> and the <strong>HTTPBin MCP server</strong> you created in Lab 2.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ONaskZdNQzIc8IHxlrydrw.png" /></figure><p><strong>NOTE: </strong>Once the synchronization is completed, any new MCP servers you add to APIM will automatically appear in API Center. This is the foundation of your private registry.</p><h4>3 — Register Partner MCP Servers</h4><p>In addition to syncing from APIM, Azure API Center includes a built-in MCP discovery page with pre-listed partner MCP servers that you can add to your private registry. These are real, production MCP servers maintained by Microsoft partners.</p><ol><li>In the left menu, under <strong>Discover</strong>, select <strong>MCP</strong>.</li></ol><p>2. Browse the available partner MCP servers. You will see available integrations like the following (unfortunately, the catalog is very light at the moment):</p><ul><li>Atlassian: Interact with Jira and Confluence through MCP tools</li><li>Playwright: Browser automation for testing web applications</li><li>Other partner servers as they are added to the program</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Zt-EwcGEWuydAxeU03ZlEw.png" /></figure><p>3. Select <strong>GitHub</strong> and click <strong>Register</strong> to add it to your inventory.</p><p>4. Follow any on-screen instructions to complete the registration.</p><p><strong>Note: </strong>When you register a partner MCP server, API Center automatically creates an API entry (with the API type set to MCP), an environment, a deployment, and an OpenAPI definition if available from the partner. This saves significant manual configuration.</p><h4>Verify the Registration</h4><p>1. Navigate to <strong>Inventory</strong> <strong>→ Assets</strong>.</p><p>2. You should now see the <strong>GitHub</strong> MCP server alongside your APIM-synced HTTPBin server — all in one inventory.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ghj7T3qSkvjJ1uq_-cQZ_A.png" /></figure><h4>4 — Deploy the API Center Portal</h4><p>The developer portal is the private “MCP app store” that teams across your organization can go to, to discover and connect to approved MCP servers.</p><h4>Task 4.1— Configure the Identity Provider</h4><p>1. In the left menu, under <strong>API Center Portal (preview)</strong>, open the <strong>Settings</strong> page.</p><p>2. In the <strong>Access</strong> tab, click <strong>Configure Entra ID</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0eUg-i-QCdkCdaTkJDImjg.png" /></figure><p>3. In the<strong> Quick setup </strong>tab, select <strong>Configure</strong>.</p><p><strong>Note: </strong>This creates a Microsoft Entra ID app registration for the portal. In production, this controls which users can access the portal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/865/1*Jn8_G5IUNPqW28B54mofTA.png" /></figure><p>4. Click <strong>Save + publish</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9joP9w7ZFJoBd7u9RS-hLg.png" /></figure><h4>Task 4.2 — Configure the Site Profile</h4><p>1. Go to the <strong>Site Profile</strong> tab.</p><p>2. Set a name for your portal (e.g., the name of your API Center resource).</p><h4>Task 4.3 — Configure Visibility</h4><p>1. Go to the <strong>Visibility</strong> tab.</p><p>2. Enable <strong>Allow anonymous access</strong>.</p><p><strong>Important: </strong>For this lab we enable anonymous access to skip additional configuration. In production, you probably will use a more secure implementation so only authorized users can discover MCP server endpoints.</p><h4>Task 4.4— Publish and Open</h4><p>1. Click <strong>Save + publish</strong> at the top of the page.</p><p>2. Click <strong>View API Center Portal</strong>.</p><p>You have successfully deployed the API Center Developer Portal!</p><h4>5 — Explore the Private MCP Registry</h4><p>You should now see all the MCP servers and APIs in your registry — the HTTPBin MCP server from Lab 2 alongside the partner MCP servers you registered.</p><h4>Task 5.1 — Browse the Catalog</h4><p>1. Browse the list of APIs and MCP servers.</p><p>2. Notice that the HTTPBin MCP server (auto-synced from APIM) appears alongside the partner servers you registered manually.</p><p>3. Use the search and filter options to narrow the list.</p><h4>Task 5.2— Copy an MCP Server Endpoint URL</h4><p>1. Click on the <strong>HTTPBin MCP</strong> entry (the API you deployed and governed in Lab 2).</p><p>2. Scroll down to <strong>Options</strong> and click <strong>Copy URL</strong> to copy the governed APIM endpoint to your clipboard.</p><p>This is the core value of the private registry: developers come to one place, find all approved MCP servers, and get the exact endpoint URLs they need. Developers who do not have direct access to the APIM resource can still discover and use these servers through the portal.</p><h4>6 — Connect to VS Code and GitHub Copilot (Optional)</h4><p>Azure API Center can serve as a private MCP registry that VS Code and GitHub Copilot consume directly. Developers’ AI tools will only discover and use the MCP servers your organization has approved.</p><h4>Task 6.1 — Get the Registry Endpoint</h4><p>Your API Center portal exposes a registry endpoint following the MCP Registry v0.1 specification:</p><p>https://&lt;your-portal&gt;.portal.azure-apicenter.ms/workspaces/default</p><p>Replace &lt;your-portal&gt; with your actual API Center portal hostname (visible in the portal URL).</p><h4>Task 6.2 — Configure VS Code</h4><p>1. In VS Code, open <strong>Settings</strong> (Ctrl+Comma / Cmd+Comma).</p><p>2. Search for <strong>mcp registry</strong>.</p><p>3. Add your API Center registry URL.</p><p>4. Restart VS Code.</p><p>GitHub Copilot will fetch the list of MCP servers from your private registry. Only approved servers will be available.</p><p><strong>Tip: </strong>Verify by opening the Output pane in VS Code and selecting the “Window” channel — you should see the registry URL being fetched.</p><h4>7 — Clean Up</h4><p>If you want to remove all resources created across Labs 2 and 3:</p><pre>az group delete --name MCPLabResources --yes --no-wait</pre><p>This deletes the App Service, APIM instance, and API Center resource.</p><h4>8 — Summary</h4><p>Across Labs 1–3, you have built a complete, end-to-end MCP management stack:</p><ul><li><a href="https://medium.com/@davidokeyode/ai-hands-on-lab-for-the-security-engineer-1-exploring-mcp-servers-public-registries-37a44fce4a47">Lab 1</a>: Explored public registries, connected to real MCP servers (Web Fetch, GitHub) using the Inspector</li><li><a href="https://davidokeyode.medium.com/ai-hands-on-lab-for-the-security-engineer-2-governing-mcp-servers-with-an-ai-gateway-in-azure-0101bd7376b3">Lab 2</a>: Deployed a REST API (httpbin), governed it through APIM with JWT auth and rate limiting, and exposed it as an MCP server</li><li>Lab 3: Built a private MCP registry with API Center for enterprise-wide discovery of real MCP servers</li></ul><h4>9 — Key Takeaways</h4><ul><li>Azure API Management governs MCP traffic (security, rate limiting, monitoring) — the runtime gateway in the data path.</li><li>Azure API Center governs MCP discovery (inventory, metadata, portal) — the organizational catalog in the control plane.</li><li>Together, they can be part of your secure AI architecture providing discovery + governance + observability.</li></ul><h4>10 — Useful Links</h4><ul><li><a href="https://mcp.azure.com/">MCP Center Live Demo — mcp.azure.com</a></li><li><a href="https://learn.microsoft.com/en-us/azure/api-center/register-discover-mcp-server">Register &amp; Discover MCP Servers in API Center — learn.microsoft.com</a></li><li><a href="https://learn.microsoft.com/en-us/azure/api-management/mcp-server-overview">APIM MCP Server Overview — learn.microsoft.com</a></li><li><a href="https://techcommunity.microsoft.com/blog/integrationsonazureblog/build-secure-launch-your-private-mcp-registry-with-azure-api-center-/4438016">Private MCP Registry Blog — Microsoft Tech Community</a></li><li><a href="https://github.com/Azure-Samples/mcp-registry">MCP Registry Sample — Azure-Samples/mcp-registry</a></li><li><a href="https://devblogs.microsoft.com/all-things-azure/locking-down-mcp-create-a-private-registry-on-azure-api-center-and-enforce-it-in-github-copilot-and-vs-code/">Locking Down MCP Guide — Microsoft DevBlogs</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f632b7e9c1fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Step-by-Step Guide: Connect AI Models in Alibaba Cloud Model Studio with OpenClaw]]></title>
            <link>https://davidokeyode.medium.com/step-by-step-guide-connect-ai-models-in-alibaba-cloud-model-studio-with-openclaw-ada70b6977cb?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/ada70b6977cb</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[openclaw]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[alibabacloud]]></category>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Sat, 21 Mar 2026 19:38:09 GMT</pubDate>
            <atom:updated>2026-03-21T19:38:09.127Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fn2rA2gURQZZQWQEokYwYA.png" /></figure><h4>Create Workspace</h4><p>Go to <a href="https://modelstudio.console.alibabacloud.com/">https://modelstudio.console.alibabacloud.com/</a> → Sign in → Settings (top right) → Workspaces → Add Workspace</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EghrWBa__Bn-jPH8LK9qqg.png" /></figure><p>In the Add Workspace menu, configure the following:</p><ul><li><strong>Workspace Name</strong>: Dev-Workspace</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/886/1*ChPQT07K5dTuAIvsljct5g.png" /></figure><p>After the workspace is created, click on <strong>Authorization &amp; Throttling Settings</strong> to configure the ____</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*20UTrMc8E-aDeowa5nVl8A.png" /></figure><p>This is where you can enable the models that will be available, configure request and token limits for them. To enable a model and configure its limits, click <strong>Edit </strong>in the <strong>Actions </strong>column</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6-REmt3oRv_GQIDbWak0xg.png" /></figure><p>In the <strong>Model Calling</strong> column, toggle the <strong>Unauthorized </strong>switch to <strong>Authorized</strong>.</p><ul><li>For the <strong>Request Number Limit</strong>, configure the <strong>Current Workspace Limit.</strong> Configure a limit that is within the stated <strong>Account Limit</strong>.</li><li>For the <strong>Token Limit</strong>, configure the <strong>Current Workspace Limit. </strong>Configure a limit that is within the stated <strong>Account Limit</strong>.</li></ul><p>Click on <strong>Save </strong>to save it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eRXMi_XBCNNc9SeBtmyCTw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O_kD5IEW4YGszkLoPhJ-Ug.png" /></figure><p>The model should now show as being <strong>Authorized</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8ge872Zlr5WHVN4wN--HUw.png" /></figure><p>Follow the above process to authorize and set limits for the following models:</p><ul><li>qwen3.5-plus</li><li>qwen3-max-2026–01–23</li><li>qwen3-coder-next</li><li>qwen3-coder-plus</li></ul><h4>Create API Key</h4><p>In the model studio <a href="https://modelstudio.console.alibabacloud.com/">https://modelstudio.console.alibabacloud.com/</a>, click on Settings (top right) → API Key → Create API Key</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zKZOEkBPajnNWeg_wXYarA.png" /></figure><p>In the <strong>Create API Key</strong> window, select the <strong>Owner Account</strong>, select the <strong>Dev-Workspace</strong>, then click <strong>OK</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WqqffdtRbLTML7H5JrQYSw.png" /></figure><p>Click the <strong>copy </strong>button to copy the <strong>API Key</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PqjWFRaaNWLCuY4nQCqRQw.png" /></figure><h4>Add a RAM Account to the Workspace and grant them permissions to the workspace</h4><p>In the model studio <a href="https://modelstudio.console.alibabacloud.com/">https://modelstudio.console.alibabacloud.com/</a>, click on Settings (top right) → Accounts → Add User (top right)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ekRqVvAUwOziqORm2Bt3Fw.png" /></figure><p>In the Add User window, configure the following:</p><ul><li><strong>Type</strong>: RAM User</li><li><strong>RAM User</strong>: Select a RAM user account</li><li><strong>Display Name</strong>: Leave the default entrance</li><li>Click <strong>OK</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/1*K5qP-AQaMlUqZoAGJSuGWg.png" /></figure><p>On the <strong>Accounts </strong>page, click <strong>Permissions</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-zyFbs8_pTz5LJf-witbwA.png" /></figure><p>Then click <strong>Add Permission</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/926/1*eplvq6Z_lDUZdXebqea0qg.png" /></figure><p>In the <strong>Edit Permission</strong> window, select the <strong>Dev-Workspace</strong>, then select <strong>Next</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MZlqylt9YjDs2kdNQShKCg.png" /></figure><p>Still in the <strong>Edit Permission</strong> window, select the permission that you want to grant the user E.g. <strong>Administrator</strong>. Then click <strong>Complete</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4KlppMQMTzqLG_kP2pq73g.png" /></figure><p>The permission should now be granted to the RAM user.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/958/1*wQZo8IT9oCacl4ZBAgBdzA.png" /></figure><h4>Configure the models in OpenClaw</h4><p>If the gateway is running, kill it first:</p><pre>tmux kill-session -t openclaw 2&gt;/dev/null<br>pkill -9 -f openclaw<br>sleep 1</pre><p>Add the following to the OpenClaw config file:</p><pre># Edit directly<br>vim ~/.openclaw/openclaw.json<br><br># Update config with both providers<br>  &quot;models&quot;: {<br>    &quot;mode&quot;: &quot;merge&quot;,<br>    &quot;providers&quot;: {<br>      &quot;bailian&quot;: {<br>        &quot;baseUrl&quot;: &quot;https://dashscope-intl.aliyuncs.com/compatible-mode/v1&quot;,<br>        &quot;apiKey&quot;: &quot;YOUR_API_KEY&quot;,<br>        &quot;api&quot;: &quot;openai-completions&quot;,<br>        &quot;models&quot;: [<br>          {<br>            &quot;id&quot;: &quot;qwen3.5-plus&quot;,<br>            &quot;name&quot;: &quot;qwen3.5-plus&quot;,<br>            &quot;reasoning&quot;: false,<br>            &quot;input&quot;: [&quot;text&quot;, &quot;image&quot;],<br>            &quot;contextWindow&quot;: 1000000,<br>            &quot;maxTokens&quot;: 65536<br>          },<br>          {<br>            &quot;id&quot;: &quot;qwen3-coder-next&quot;,<br>            &quot;name&quot;: &quot;qwen3-coder-next&quot;,<br>            &quot;reasoning&quot;: false,<br>            &quot;input&quot;: [&quot;text&quot;],<br>            &quot;contextWindow&quot;: 262144,<br>            &quot;maxTokens&quot;: 65536<br>          }<br>        ]<br>      }<br>    }<br>  },<br>  &quot;agents&quot;: {<br>    &quot;defaults&quot;: {<br>      &quot;model&quot;: {<br>        &quot;primary&quot;: &quot;bailian/qwen3.5-plus&quot;<br>      },<br>      &quot;models&quot;: {<br>        &quot;bailian/qwen3.5-plus&quot;: {},<br>        &quot;bailian/qwen3-coder-next&quot;: {}<br>      }<br>    }<br>  },</pre><p>Restart the gateway (in the background):</p><pre># Start gateway in background<br>tmux new -d -s openclaw &#39;openclaw gateway run&#39;<br><br># Verify<br>sleep 3<br>tmux ls<br>openclaw doctor</pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ada70b6977cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Hands-On Lab for the Security Engineer (2)— Governing MCP Servers with an AI Gateway in Azure]]></title>
            <link>https://davidokeyode.medium.com/ai-hands-on-lab-for-the-security-engineer-2-governing-mcp-servers-with-an-ai-gateway-in-azure-0101bd7376b3?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/0101bd7376b3</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai-security]]></category>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Sun, 08 Mar 2026 00:00:28 GMT</pubDate>
            <atom:updated>2026-03-08T00:00:28.737Z</atom:updated>
            <content:encoded><![CDATA[<p>Estimated Duration: 60–75 minutes | Difficulty: Intermediate</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DLBUgA8T8zHyKB7ZrmvQDQ.png" /></figure><h3>Background</h3><p>In <a href="https://davidokeyode.medium.com/ai-hands-on-lab-for-the-security-engineer-1-exploring-mcp-servers-public-registries-37a44fce4a47">Lab 1</a> you explored public MCP registries, launched the MCP Inspector, and connected directly to real MCP servers (Web Fetch and GitHub). Those connections were point-to-point, where your MCP client talked straight to the server with no intermediary.</p><p>That approach works for experimentation, but in an enterprise environment it creates serious challenges: no centralized authentication, no rate limiting, no monitoring, and no way to enforce security policies across all the MCP servers your organization uses. To implement these centralized governance capabilities, it is recommended to implement an AI gateway.</p><p>In this lab you will <strong>deploy a REST API as a container on Azure App Service</strong>, import it into Azure API Management (APIM), and expose its operations as MCP tools, all governed by APIM’s security policies. This walks you through how enterprises can turn existing production APIs into agent-callable MCP tools.</p><h3>What You Will Build</h3><ul><li>Deploy the httpbin API as a Linux container on Azure App Service.</li><li>Import the httpbin API into Azure API Management (APIM) and define operations.</li><li>Expose those API operations as MCP server tools that AI agents can call.</li><li>Apply JSON Web Token (JWT) validation and rate-limiting policies to govern all tool calls.</li><li>Test the governed MCP server end-to-end using the MCP Inspector.</li></ul><h3>Why Govern MCP Servers?</h3><p>Without governance:</p><ul><li>Credential/Authentication management is fragmented across individual MCP server deployments. Some of them may even be exposed without authentication. This is not a good security practice.</li><li>There is no visibility into which agents are calling which tools, or how often.</li><li>Malicious or misconfigured agents can overwhelm backend services with unlimited requests.</li></ul><h3>Understanding the Two Azure Services</h3><p>Azure provides two complementary services for MCP management. This lab covers the first; In the next hands-on-lab after this, we will cover the second service. The way to think about it is that <strong>APIM governs the client to server connection</strong> (this lab), while <strong>API Center governs the enterprise discover </strong>(Think about this as an internal “app store” for available MCP servers. We’ll cover this in Lab 3).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/887/1*_GYON-2xub3X5U7pssuDEQ.png" /></figure><h3>Prerequisites</h3><ul><li>Completed Lab 1 (familiarity with MCP Inspector and MCP concepts).</li><li>An Azure subscription with at least Contributor access.</li><li>Azure CLI installed and authenticated using <strong>az login</strong> .</li><li>Node.js v18+ installed locally</li><li>A modern web browser</li></ul><h3>Exercise 1 — Deploy a REST API to Azure App Service</h3><p>First, you will deploy a REST API as a container to Azure App Service. We will use httpbin — a widely-used HTTP request and response testing service created by <a href="https://github.com/kennethreitz">Kenneth Reitz</a>. It provides dozens of real endpoints for testing HTTP methods, status codes, authentication, headers, and more.</p><p>This will represent an API service that an enterprise would already have running. Our objective in this lab is to turn it into an MCP server so AI agents can call its endpoints as MCP tools.</p><h4>Task 1.1 — Configure Variables</h4><p>Open a terminal (Bash, PowerShell, or Azure Cloud Shell) and set the following variables:</p><pre>appName=&quot;httpapi$RANDOM&quot;<br>location=&quot;eastus&quot;<br>appPlan=&quot;$location-lin-plan&quot;<br>group=&quot;MCPLabResources&quot;</pre><h4>Task 1.2 — Create the Resource Group</h4><pre>az group create -l $location -n $group</pre><h4>Task 1.3 — Create the App Service Plan</h4><p>Linux containers can run on Standard, Pv2, or Pv3 Linux plans:</p><pre>az appservice plan create -n $appPlan -g $group -l $location --sku B1 --number-of-workers 1 --is-linux</pre><h4>Task 1.4 — Deploy the httpbin Container</h4><p>Deploy the httpbin container image from <a href="https://hub.docker.com/r/kennethreitz/httpbin">Docker Hub</a>:</p><pre>image=&quot;kennethreitz/httpbin:latest&quot;<br>az webapp create -n $appName -g $group --plan $appPlan --deployment-container-image-name $image</pre><h4>Task 1.5 — Verify the Deployment</h4><p>1. Get the URL of your deployed web app:</p><pre>echo &quot;https://$appName.azurewebsites.net&quot;</pre><p><strong>Tip: </strong>Record your app URL (https://&lt;appName&gt;.azurewebsites.net). You will need it in the next step.</p><p>2. Open the URL in your browser. You should see the httpbin Swagger UI with all available endpoints.</p><p><strong>Note: </strong>There may be a 2–3 minute delay before the container is fully started. If you get a 504 Gateway Timeout, wait and refresh.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*HvvA2TmH5WBaqiDLkgXYtA.png" /></figure><p>3. Test a few endpoints from the terminal:</p><pre>curl -X GET &quot;https://$appName.azurewebsites.net/html&quot; -H &quot;accept: text/html&quot;<br>curl -X GET &quot;https://$appName.azurewebsites.net/ip&quot; -H &quot;accept: application/json&quot;<br>curl -X GET &quot;https://$appName.azurewebsites.net/user-agent&quot; -H &quot;accept: application/json&quot;</pre><h3>2 — Create Azure API Management and Import the API</h3><h4>Task 2.1 — Create an APIM Instance</h4><p>If you do not already have an APIM instance, create one now.</p><p><strong>Important: </strong>The Consumption tier does NOT support MCP servers. MCP servers are supported in all tiers except Consumption. In this lab, we will use the <strong>Developer </strong>tier.</p><p>Before running the command, replace the value of the --publisher-email with your email.</p><pre>apimName=&quot;mcplab-apim-$RANDOM&quot;<br>az apim create -n $apimName -g $group --publisher-name &quot;MCP Lab&quot; --publisher-email YOUR_EMAIL_ADDRESS --sku-name Developer --no-wait</pre><p><strong>Note: </strong>The Developer tier can take 30–40 minutes to provision. You can also create the instance via the Azure Portal. Adding --no-wait to the command ensures the command line is not blocked.</p><h4>Task 2.2 — Import the httpbin API Using Its Swagger Specification</h4><ol><li>In the Azure Portal, wait for the status of the resource to change from Activating to Online.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V5N2VCZD9BiB16XMXq22WQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OBelDeRMX60zi4EDBxjcVg.png" /></figure><blockquote>httpbin exposes a Swagger (OpenAPI) specification file that APIM can import directly — no need to manually define operations one by one.</blockquote><p>2. Select your APIM resource. In the left menu, under <strong>APIs</strong>, select <strong>APIs</strong>.</p><p>3. In the <strong>Create from definition</strong> section, select <strong>OpenAPI</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CSmlvs2Yj7Rux5qHEahSBA.png" /></figure><p>4. In the import dialog, select the Full option and fill in:</p><ul><li><strong>OpenAPI specification</strong>: https://&lt;your-app-name&gt;.azurewebsites.net/spec.json (Replace &lt;your-app-name&gt; with the app name in the URL from Task 1.5)</li><li><strong>OpenAPI specification</strong>: https://raw.githubusercontent.com/Azure/api-management-samples/master/apis/httpbin.swagger.json</li><li><strong>Display name</strong>: HTTPBin API</li><li><strong>Name</strong>: httpbin-api</li><li><strong>URL scheme</strong>: HTTPS</li><li><strong>API URL suffix</strong>: httpbin</li></ul><p>5. Click <strong>Create</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tP7BBNsac0HnQ5iRTPXeaA.png" /></figure><p>APIM imports all the httpbin operations automatically from the Swagger file — including endpoints for HTTP methods, status codes, authentication testing, request inspection, response formats, dynamic data, and more. Each operation has proper names, descriptions, and parameter schemas.</p><h4>Task 2.3 — Update the Web Service URL</h4><p>The Swagger file has <a href="https://httpbin.org/"><strong>https://httpbin.org</strong></a> hardcoded as the host. You need to change it to point to <strong>your own deployed container</strong> on App Service.</p><p><strong>Important: </strong>These steps are critical. Without them, APIM will proxy requests to the public httpbin.org instead of your own deployed container and will require a subscription key when testing. In a real enterprise scenario, you always want traffic routed to your own backend, not a third-party service.</p><p>1. On the imported HTTPBin API, click the <strong>Settings</strong> tab.</p><p>2. In the <strong>Web service URL</strong> field, replace https://httpbin.org with your App Service URL: https://&lt;your-app-name&gt;.azurewebsites.net (Replace &lt;your-app-name&gt; with the app name in the URL from Task 1.5)</p><p>3. Scroll down to the <strong>Subscription</strong> section and uncheck “<strong>Subscription required”</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WqhU72zdU2BXPpQhp7uJwQ.png" /></figure><p>4. Click <strong>Save</strong>.</p><h3>3 — Expose the API as an MCP Server</h3><p>Now you will turn the httpbin REST API into an MCP server — making each operation callable as an MCP tool by AI agents.</p><p>1. In the left menu of your APIM instance, under <strong>APIs</strong>, select <strong>MCP Servers</strong>.</p><p>2. Click <strong>+ Create MCP server</strong> and select <strong>Expose an API as an MCP server</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NlYyDOhdJR_XRNB83r5IOg.png" /></figure><p>3. Fill in the form:</p><ul><li><strong>API</strong>: Select HTTPBin API</li><li><strong>API Operations</strong>: Select the operations you want to expose as MCP tools (e.g., [GET] /get, [GET] /headers, [GET] /ip, [GET] /status, [GET] /user-agent, and others)</li><li><strong>Display name</strong>: HTTPBin MCP</li><li><strong>Name</strong>: httpbin-mcp</li><li><strong>Products</strong>: Leave empty</li></ul><p><strong>Tip: </strong>The Swagger import gave you many operations to choose from. Select 4–6 that make sense as agent tools — for example, IP lookup, header inspection, status code testing, and HTML retrieval. You can always add more later.</p><p>4. Click <strong>Create</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qvda_9w9s0H_FLV9eZiV9w.png" /></figure><p>You have now turned a REST API into an MCP server within APIM — <strong>without writing any backend code!</strong> The Swagger-imported operations are now exposed as MCP tools that any AI agent can call through the MCP protocol.</p><p><strong>NOTE: </strong>Copy the Server URL shown in the MCP Servers list — you will need it for testing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*idwTsOGhK6UvYNb-2mbC3w.png" /></figure><h3>4 — Apply Security Policies</h3><p>Next, you will add some governance. You will add a JWT validation policy so that only authenticated users can call the MCP tools, and a rate limit to prevent abuse.</p><h4>Task 4.1 — Open the Policy Editor</h4><p>1. In the MCP Servers list, click on <strong>httpbin-mcp</strong>.</p><p>2. In the left menu, expand <strong>MCP</strong>, then select <strong>Policies</strong>.</p><h4>Task 4.2 — Add JWT Validation and Rate Limiting</h4><ol><li>Replace the entire policy XML with the following:</li></ol><pre>&lt;policies&gt;<br>    &lt;inbound&gt;<br>        &lt;base /&gt;<br>        &lt;cors allow-credentials=&quot;true&quot;&gt;<br>            &lt;allowed-origins&gt;<br>                &lt;origin&gt;http://localhost:6274&lt;/origin&gt;<br>            &lt;/allowed-origins&gt;<br>            &lt;allowed-methods&gt;<br>                &lt;method&gt;*&lt;/method&gt;<br>            &lt;/allowed-methods&gt;<br>            &lt;allowed-headers&gt;<br>                &lt;header&gt;*&lt;/header&gt;<br>            &lt;/allowed-headers&gt;<br>        &lt;/cors&gt;<br>        &lt;validate-jwt header-name=&quot;Authorization&quot; failed-validation-error-message=&quot;Unauthorized. Access token is missing or invalid.&quot;&gt;<br>            &lt;openid-config url=&quot;https://login.microsoftonline.com/common/.well-known/openid-configuration&quot; /&gt;<br>            &lt;audiences&gt;<br>                &lt;audience&gt;https://azure-api.net/authorization-manager&lt;/audience&gt;<br>            &lt;/audiences&gt;<br>        &lt;/validate-jwt&gt;<br>        &lt;rate-limit-by-key calls=&quot;10&quot; renewal-period=&quot;60&quot; counter-key=&quot;@(context.Request.IpAddress)&quot; /&gt;<br>    &lt;/inbound&gt;<br>    &lt;backend&gt;<br>        &lt;base /&gt;<br>    &lt;/backend&gt;<br>    &lt;outbound&gt;<br>        &lt;base /&gt;<br>    &lt;/outbound&gt;<br>    &lt;on-error&gt;<br>        &lt;base /&gt;<br>    &lt;/on-error&gt;<br>&lt;/policies&gt;</pre><p><strong>Note: </strong>The CORS origin is set to <a href="http://localhost:6274,">http://localhost:6274,</a> which is the MCP Inspector’s default port. If your Inspector started on a different port (visible in the terminal output, e.g., “Starting MCP Inspector on port 6275”), update the origin in the policy to match before saving.</p><p>2. Click <strong>Save</strong>.</p><p>This policy applies three layers of governance to every MCP tool call:</p><ul><li>cors — Allows the MCP Inspector (running on localhost:6274) to make cross-origin requests to APIM. The allow-credentials setting ensures the Authorization header is forwarded. In production, replace the localhost origin with your actual MCP client origins.</li><li>validate-jwt — Requires a valid JWT token issued by Microsoft Entra ID. Only authenticated users with a valid token can call the MCP tools. Unauthenticated requests are rejected with a 401 Unauthorized error.</li><li>rate-limit-by-key — Limits each client IP address to 10 tool calls per 60 seconds, preventing any single agent from overwhelming the backend.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vaouvxofUoZpokZGi4m-gg.png" /></figure><h4>Other Policies You Can Apply</h4><ul><li>ip-filter — Restrict access to specific corporate IP ranges</li><li>set-header — Inject backend API keys automatically so clients never see them</li><li>log-to-eventhub — Stream MCP traffic to Event Hubs for auditing</li><li>cache-lookup/cache-store — Cache frequent tool responses to reduce backend load</li></ul><h3>5 — Test the Governed MCP Server</h3><h4>Task 5.1 — Get an Access Token</h4><ol><li>Open the Azure Cloud Shell (Bash) or a terminal with Azure CLI and run the following.</li></ol><p><strong>Note: </strong>Even if you’re using Azure Cloud Shell, you must still run az login.</p><pre>az login<br>az account get-access-token --resource &quot;https://azure-api.net/authorization-manager&quot; --query accessToken --output tsv</pre><p>2. Copy the returned token to your clipboard.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/910/1*bLTwaic0KPJjp92tkRV7Mg.png" /></figure><p><strong>Tip: </strong>If you get an error, ensure you have run az login and your account has access to the APIM instance.</p><h4>Task 5.2 — Launch the MCP Inspector</h4><p>1. In a new terminal, launch:</p><pre>npx -y @modelcontextprotocol/inspector</pre><p>2. Wait for the browser to open. If it does not, click the URL in the terminal output.</p><h4>Task 5.3— Connect to Your Governed Endpoint</h4><p>1. Set <strong>Transport type</strong> to <strong>Streamable HTTP</strong>.</p><p>2. Paste the <strong>Server URL</strong> from APIM’s MCP Servers list into the URL field.</p><p>3. Set <strong>Connection Type</strong> to <strong>Via Proxy</strong>.</p><p>4. Under <strong>Authentication</strong>, expand the section and click <strong>+ Add</strong> to add a custom header.</p><p>5. Configure the header with these exact values:</p><ul><li><strong>Header name</strong>: Authorization</li><li><strong>Header value</strong>: Type: <strong>Bearer </strong>(with a trailing space) then paste your access token.</li></ul><p>6. Make sure the toggle next to the header is switched ON (enabled).</p><p>7. Click <strong>Connect</strong>. You should see a green <strong>Connected</strong> label.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/860/1*b8n0w5nEf_lFoSew_WFn6A.png" /></figure><p>6. Click <strong>Connect</strong>. You should see a green <strong>Connected</strong> label.</p><p><strong>⚠️ Important: </strong>Ensure the Authorization header toggle is enabled and the value starts with “Bearer” (with a space). Without these, authentication will fail.</p><h4>Task 5.4 — Invoke the httpbin Tools</h4><p>1. Click the <strong>Tools</strong> tab, then <strong>List Tools</strong>.</p><p>2. You should see the five tools corresponding to your httpbin operations: [GET] /get, [GET] /headers, [GET] /ip, [GET] /status, [GET] /user-agent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*Y85vieeiBs-vIoaNvsYQHw.png" /></figure><p>3. Click on <strong>ip</strong> and click <strong>Run Tool</strong>.</p><p>4. Review the result — you should see a JSON response with the origin IP address. This response came from the live httpbin container, proxied through APIM.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/993/1*Bfk2OJFUtCQ_lBIe_pJAJg.png" /></figure><p>5. Try the <strong>headers</strong> tool— this shows all HTTP headers the backend received, including any headers APIM injected via policies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/978/1*fD_1yLWSW65Hlrm_qR8kgg.png" /></figure><p>6. Try <strong>userAgent </strong>tool — this shows the user-agent string, revealing how the MCP Inspector identifies itself to backends.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*lYzBbGqShkKatSiLW-PPpA.png" /></figure><p>You have just used the MCP Inspector to call tools on a real API (httpbin) that is running as a container on Azure App Service, governed by Azure API Management with JWT authentication and rate limiting.</p><h4>Task 5.5— Verify Security</h4><p>1. Remove or corrupt the Bearer token in the Authorization header.</p><p>2. Try to Connect or invoke a tool.</p><p>3. You should receive an Unauthorized error — confirming the JWT policy blocks unauthenticated requests.</p><h3>6 — Clean Up</h3><p><strong>Important: </strong>This will delete the entire resource group including the App Service, App Service Plan, and APIM instance. <strong>Skip this for now if you plan to continue with Lab 3.</strong></p><p>If you want to remove the resources created in this lab:</p><pre>az group delete — name MCPLabResources — yes — no-wait</pre><h3>7 — Summary</h3><p>In this lab you have:</p><ul><li>Deployed a REST API (httpbin) as a Linux container on Azure App Service</li><li>Imported the API into Azure API Management and defined operations</li><li>Exposed those operations as MCP server tools — turning a production API into agent-callable tools without backend code changes</li><li>Applied JWT validation and per-session rate limiting to govern all MCP tool calls</li><li>Tested the end-to-end governed flow using the MCP Inspector with a Bearer token</li></ul><h3>8— Key Takeaways</h3><ul><li>Azure API Management is an enterprise runtime gateway for MCP servers. It proxies every tool call and enforces policies in real time.</li><li>Any existing REST API managed in APIM can be exposed as an MCP server with zero backend code changes.</li><li>APIM’s full policy engine (JWT, rate limits, IP filtering, CORS, caching, logging) applies to all MCP tool calls.</li><li>APIM currently supports MCP tools (not resources or prompts). External servers must use MCP spec 2025–06–18+ with Streamable HTTP or SSE.</li></ul><p>In <strong>Lab 3</strong>, you will use <strong>Azure API Center</strong> to build a private MCP registry. This will enable enterprise-wide discovery of your governed MCP servers and preventing “shadow AI”.</p><h3>9 — Useful Links</h3><ul><li><a href="https://learn.microsoft.com/en-us/azure/api-management/mcp-server-overview">APIM MCP Overview — learn.microsoft.com</a></li><li><a href="https://learn.microsoft.com/en-us/azure/api-management/export-rest-mcp-server">Expose REST API as MCP Server — learn.microsoft.com</a></li><li><a href="https://learn.microsoft.com/en-us/azure/api-management/expose-existing-mcp-server">Expose Existing MCP Server — learn.microsoft.com</a></li><li><a href="https://learn.microsoft.com/en-us/azure/api-management/genai-gateway-capabilities">AI Gateway Capabilities — learn.microsoft.com</a></li><li><a href="https://github.com/modelcontextprotocol/inspector">MCP Inspector — GitHub</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0101bd7376b3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Hands-On Lab for the Security Engineer (1) — Exploring MCP Servers & Public Registries]]></title>
            <link>https://davidokeyode.medium.com/ai-hands-on-lab-for-the-security-engineer-1-exploring-mcp-servers-public-registries-37a44fce4a47?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/37a44fce4a47</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Sat, 28 Feb 2026 23:25:34 GMT</pubDate>
            <atom:updated>2026-03-03T18:53:07.727Z</atom:updated>
            <content:encoded><![CDATA[<p>Estimated Duration: 30–45 minutes | Difficulty: Beginner</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4_vshxEgEvYf1cyZGT0bEg.jpeg" /></figure><h3>Prerequisites</h3><ul><li>A computer with Node.js (v18 or higher) installed</li></ul><pre>node --version</pre><ul><li>A modern web browser (Chrome, Edge, or Firefox)</li><li>A terminal / command prompt</li><li>Internet access</li><li>A GitHub account</li></ul><h3>1 — What Is an MCP Server?</h3><p>The Model Context Protocol (MCP) is an open standard that defines how AI applications communicate with external data sources, tools, and services (servers). An MCP server <strong>exposes three main components </strong>to clients:</p><ul><li><strong>Tools </strong>— These are functions that the AI app can call (e.g. search documents, add numbers, fetch a webpage).</li><li><strong>Resources </strong>— These are structured data the AI can read (e.g. files, database records, configuration).</li><li><strong>Prompts</strong> —These are pre-built prompt templates the AI can use for common tasks.</li></ul><p>MCP servers can run locally on a machine (<strong><em>via STDIO transport</em></strong>) or remotely over the network (<strong><em>via Streamable HTTP or SSE transport</em></strong>). In this lab, you will run a local test server and connect to it using the<strong> MCP Inspector</strong>, a browser-based developer tool.</p><h3>2 — Explore Public MCP Registries</h3><p>Let’s start the hands-on by exploring where MCP servers are discovered. Several public registries and directories publish official and community created MCP servers (similar to the concept of an app store).</p><h4>Exercise 2.1 — The Official MCP Registry</h4><p>1. Open your browser and navigate to the Official MCP Registry:</p><p><a href="https://registry.modelcontextprotocol.io/">https://registry.modelcontextprotocol.io</a></p><p>2. Browse the list of servers. You will see a searchable catalog of publicly registered MCP servers, including their names, descriptions, and metadata.</p><p>3. Use the search box to search for: <strong><em>filesystem</em></strong></p><p>4. Click on one of the results to view its details. Notice the metadata fields: package name, version, description, and how to install it.</p><p><strong>NOTE: </strong>The Official MCP Registry launched in September 2025. It is maintained by the MCP open-source community with backing from popular providers like Anthropic, GitHub, Microsoft, and PulseMCP.</p><h4>Exercise 2.2 — The GitHub MCP Organization</h4><p>1. Navigate to the GitHub MCP organization page:</p><p><a href="https://github.com/mcp">https://github.com/mcp</a></p><p>2. Scroll through the list of servers. These are official and community-contributed servers hosted directly under the GitHub MCP registry.</p><p>3. Click on a server that interests you (e.g. the GitHub server, Stripe server, or Playwright server) and read its README to understand what tools it provides.</p><p><strong>NOTE: </strong>The GitHub MCP organization centralizes MCP servers for easy discovery and integration. Each server includes installation instructions and a list of tools it exposes.</p><h4>Exercise 2.3 — PulseMCP Directory</h4><p>1. Navigate to the PulseMCP server directory:</p><p><a href="https://www.pulsemcp.com/servers">https://www.pulsemcp.com/servers</a></p><p>2. Explore the filters at the top. You can filter by classification (Official, Community, Anthropic References) and sort by popularity, recency, or alphabetically.</p><p>3. Toggle the “<strong>Remote Available</strong>” filter to see which servers can run remotely over the network.</p><p>4. Pick any server and note its description. We will reference this later.</p><h4>Exercise 2.4 — Query the Registry API (Optional)</h4><p>The Official MCP Registry also exposes a REST API. Try running this command in your terminal:</p><pre># Bash<br>curl &quot;https://registry.modelcontextprotocol.io/v0/servers?limit=5&quot;<br><br># PowerShell<br>Invoke-RestMethod &quot;https://registry.modelcontextprotocol.io/v0/servers?limit=5&quot; | ConvertTo-Json -Depth 10</pre><p>You should receive a JSON response listing five MCP servers with their metadata. You can also test the search API:</p><pre># Bash<br>curl &quot;https://registry.modelcontextprotocol.io/v0/servers?search=github&quot;<br><br># PowerShell<br>Invoke-RestMethod &quot;https://registry.modelcontextprotocol.io/v0/servers?limit=5&quot; | ConvertTo-Json -Depth 10</pre><p><strong>NOTE: </strong>This API allows for programmatic operations. It follows the MCP Registry OpenAPI specification.</p><h3>3 — Launch the MCP Inspector</h3><p>The MCP Inspector is an interactive developer tool for testing and debugging MCP servers. It provides a browser-based UI that lets you connect to any MCP server, list its tools/resources/prompts, and invoke them — all without writing code.</p><p>1. Open a terminal window.</p><p>2. Run the following command to launch the MCP Inspector:</p><pre>npx -y @modelcontextprotocol/inspector</pre><p>3. If prompted to install the package, type <strong>y</strong> and press Enter.</p><p>4. Wait for the Inspector to start.</p><p>5. A browser window should open automatically. If not, click the link shown in the terminal (the one that includes the MCP_PROXY_AUTH_TOKEN parameter for auto-authentication).</p><p><strong>IMPORTANT: </strong>The Inspector generates a unique session token each time it starts. Use the URL with the token pre-filled to avoid manual entry. Do not share this token.</p><p>You should now see the MCP Inspector interface in your browser with a connection panel on the left side.</p><h3>4 — Connect to the Web Fetch MCP Server</h3><p>The first MCP server that we will test can be used by AI models to the live internet. The Web Fetch server lets AI assistants retrieve and convert web page content into clean formats (Markdown, JSON, plain text, raw HTML). No API keys are required.</p><p>For example, an AI assistant can use its capabilities to “read” a web article, pull data from a public API, or check a website. Everything powered by the MCP protocol.</p><h4>About the Web Fetch Server</h4><p>Here are the tools this server exposes:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/826/1*AekLh4x3Kw1kFB97YAkelw.png" /></figure><h4>Exercise 4.1 — Configure the Connection</h4><p>1. In the MCP Inspector, make sure the Transport Type is set to STDIO (the default).</p><p>2. In the <strong>Command</strong> field, enter:</p><pre>npx</pre><p>3. In the <strong>Arguments</strong> field, enter:</p><pre>-y @tokenizin/mcp-npx-fetch</pre><p>4. Click the <strong>Connect</strong> button.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/460/1*EjGN7y54joVAoDsogTS5VA.png" /></figure><p>After a few seconds, you should see a green “Connected” label. The Fetch server is now running locally and ready to retrieve live web content.</p><p><strong>NOTE: </strong>If connection fails, make sure Node.js is installed and you have internet access (npx downloads the package the first time).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sRYc5p2RiHucc-MNOyNtXw.png" /></figure><h3>5 — Fetch Live Web Content</h3><p>Now let’s use the Fetch server to pull real content from the internet, just like an AI assistant would.</p><p>1. Click on the Tools tab in the Inspector.</p><p>2. Click <strong>List Tools</strong>. You should see the four tools: fetch_html , fetch_json, fetch_txt, fetch_markdown.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*15XGSLS-mM3tH_URQ8N1lg.png" /></figure><h4>Exercise 5.1 — Fetch a Web Page as Markdown</h4><p>1. Click on the fetch_markdown tool.</p><p>2. In the <strong>url</strong> field, enter: https://modelcontextprotocol.io</p><p>3. Click Run Tool.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/979/1*ooa8JOPDyICXyPbVTnvv9A.png" /></figure><p>4. Review the result. You should see the MCP homepage content converted to Markdown. This tool can be used by an AI assistant to “read” a web page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/945/1*-7tIAKxp_w1K_VH3NTd3Lw.png" /></figure><h4>Exercise 5.2 — Fetch a Public REST API as JSON</h4><p>1. Click on the fetch_json tool.</p><p>2. In the <strong>url</strong> field, enter this public API endpoint: https://api.github.com/repos/modelcontextprotocol/servers</p><p>3. Click Run Tool.</p><p>4. Review the JSON result. You should see repository metadata from the GitHub API. Including the name, description, star count, language, and more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/939/1*2CZrCBKfSpUUT1KQCW2fmg.png" /></figure><p><strong>NOTE: </strong>You just used an MCP tool to call a live REST API and retrieve structured data. This same capability can be used by AI agents to pull real-time information from external services.</p><h4>Exercise 5.3 — Fetch Plain Text from a News Site</h4><p>1. Click on the fetch_txt tool.</p><p>2. In the <strong>url</strong> field, enter any public website URL, for example: https://en.wikipedia.org/wiki/Model_Context_Protocol</p><p>3. Click Run Tool.</p><p>4. The result should show the Wikipedia article as clean plain text with all HTML tags removed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/937/1*ur1EfIziNn6ssEZwZqqwZA.png" /></figure><h4>Exercise 5.4 — Fetch Raw HTML (Optional)</h4><p>1. Click on the fetch_html tool and try fetching a page.</p><p>2. Compare the raw HTML output with the Markdown and plain text versions you saw earlier. <strong><em>Notice how each tool transforms the same web page into a different format optimized for different use cases.</em></strong></p><blockquote><strong>NOTE: </strong>The Fetch server demonstrates a key MCP pattern. The AI client/agent never deals directly with HTTP requests, HTML parsing, or content extraction. It calls the tools offereb by the MCP server to handles that.</blockquote><h3>6 — Connect to the GitHub MCP Server</h3><p>Now let’s connect to an external platform: <strong>GitHub</strong> using the official GitHub MCP server that is built and maintained by GitHub staff. This can be used by AI assistants to search repositories, read code, manage issues, create pull requests, and much more. This is a widely-used MCP server.</p><h4>Exercise 6.1— Create a GitHub Personal Access Token</h4><p>The GitHub server needs a Personal Access Token (PAT) to authenticate API requests on your behalf.</p><p>1. In your browser, navigate to: https://github.com/settings/tokens</p><p>2. Click <strong>Generate new token</strong> and select <strong>Generate new token (classic)</strong>.</p><p>3. Give the token a name, e.g. “MCP Lab”. You can set the expiration to 30 days.</p><p>4. Under <strong>Select scopes</strong>, check the following:</p><ul><li>public_repo— Grants acces to publicrepositories in the account</li><li>read:user — Grants access to read all user profile data</li></ul><p>5. Click <strong>Generate token</strong> and <strong>copy the token immediately</strong>. GitHub will only show it once.</p><p><strong>IMPORTANT: </strong>Treat your token like a password. Do not share it or commit it to source control. You can delete it after this lab from the same settings page.</p><h4>Exercise 6.2— Connect via the MCP Inspector</h4><p>1. Return to the MCP Inspector. If still connected to the Fetch server, click <strong>Disconnect </strong>first.</p><p>2. Keep the Transport Type as STDIO.</p><p>3. In the <strong>Command</strong> field, enter:</p><pre>npx</pre><p>4. In the <strong>Arguments</strong> field, enter:</p><pre>-y @modelcontextprotocol/server-github</pre><p>5. In the <strong>Environment Variables</strong> section, add a new variable:</p><ul><li><strong>Name</strong>: GITHUB_PERSONAL_ACCESS_TOKEN</li><li><strong>Value</strong>: (paste your token from the previous exercise)</li></ul><p>6. Click <strong>Connect</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/447/1*D8P-5ONbTUYr_RBypjfQ7w.png" /></figure><p>After a few seconds you should see the green “Connected” label. You are now connected to the GitHub API through MCP.</p><h3>7 — Explore GitHub Through MCP Tools</h3><p>The GitHub MCP server exposes a large set of tools. Let’s explore some common ones.</p><p>1. Click on the Tools tab, then click “<strong>List Tools</strong>”. You may need to click the Clear button before the “<strong>List Tools</strong>” button will become available again.</p><p>2. Scroll through the list. You should see multiple tools displayed.</p><h4>Exercise 7.1 — Search for Repositories</h4><p>1. Find and click on the search_repositories tool.</p><p>2. In the <strong>query</strong> field, enter: topic:mcp language:typescript stars:&gt;100</p><p>3. Click Run Tool.</p><p>4. Review the results. You should see popular TypeScript repositories related to MCP, with names, descriptions, and star counts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/953/1*T81-l8QURZwnrmnsx_Mw4Q.png" /></figure><h4>Exercise 7.2 — Get Repository Details</h4><p>1. Find and click the get_file_contents tool (or similar).</p><p>2. In the <strong>owner</strong> field, enter: modelcontextprotocol</p><p>3. In the <strong>repo</strong> field, enter: servers</p><p>4. In the <strong>path</strong> field, enter: src</p><p>4. Click <strong>Run </strong>Tool.</p><p>5. Review the result. You should see detailed metadata about the contents of the requested folder.</p><h4>Exercise 7.3 — Search Code Across GitHub</h4><p>1. Find and click the search_code tool.</p><p>2. In the <strong>query</strong> field, enter: McpServer language:typescript</p><p>3. Click Run Tool.</p><p>4. You should see results from across GitHub on code that reference the McpServer class — showing how other developers build MCP servers.</p><h4>Exercise 7.4 — List Issues on a Public Repo</h4><p>1. Find and click the list_issues<strong> </strong>tool.</p><p>2. Set <strong>owner</strong> to modelcontextprotocol and <strong>repo</strong> to servers.</p><p>3. Click Run Tool.</p><p>4. Review the open issues. You are now reading live data from a real GitHub repository through the MCP protocol.</p><p><strong>NOTE: </strong>Every tool you just used calls the GitHub REST API behind the scenes. When Claude, Copilot or your custom AI agent searches code or creates an issue, they can use this exact same MCP server and these same tools.</p><h3>8 — Clean Up</h3><p>1. Return to your terminal and press Ctrl+C to stop the MCP Inspector.</p><p>2. Close the browser tab.</p><p>3. (Recommended) Delete the GitHub Personal Access Token you created:</p><ul><li>Go to <a href="https://github.com/settings/tokens">https://github.com/settings/tokens</a></li><li>Find your “<strong>MCP Lab</strong>” token and click <strong>Delete</strong></li></ul><h3>9 — Lab Summary</h3><p>Congratulations! In this lab you have:</p><ul><li>Learned what an MCP server is and how it exposes tools, resources, and prompts to AI clients.</li><li>Explored three public MCP registries: the Official MCP Registry, GitHub MCP, and PulseMCP.</li><li>Launched the MCP Inspector and connected to real MCP servers.</li><li>Used the Web Fetch MCP server to retrieve live web pages, call a public REST API, and convert content to Markdown, JSON, and plain text.</li><li>Used the official GitHub MCP server to search repositories, inspect repo details, search code, and list issues on a real GitHub project.</li></ul><h3>10 — Key Takeaways</h3><ul><li>MCP is an open standard for connecting AI assistants to external tools and data sources.</li><li>Public registries make it easy to discover and integrate MCP servers.</li><li>The MCP Inspector is an essential developer tool for testing and debugging servers.</li><li>MCP servers wrap external services (the web, GitHub, databases, etc.) behind a consistent tool interface that any AI client can use.</li><li>Authentication (like GitHub PATs) is handled through environment variables, keeping credentials separate from configuration.</li></ul><h3>11 — Useful Links</h3><ul><li><a href="https://modelcontextprotocol.io/">MCP Official Documentation — modelcontextprotocol.io</a></li><li><a href="https://registry.modelcontextprotocol.io/">MCP Registry — registry.modelcontextprotocol.io</a></li><li><a href="https://github.com/modelcontextprotocol/servers">GitHub MCP Servers Repository — github.com/modelcontextprotocol/servers</a></li><li><a href="https://mcpmarket.com/">MCP Market — https://mcpmarket.com/</a></li><li><a href="https://www.pulsemcp.com/servers">PulseMCP Directory — pulsemcp.com/servers</a></li><li><a href="https://github.com/modelcontextprotocol/inspector">MCP Inspector — github.com/modelcontextprotocol/inspector</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=37a44fce4a47" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[OpenClaw on Azure VM — Setup Guide]]></title>
            <link>https://davidokeyode.medium.com/openclaw-on-azure-vm-setup-guide-4f0a67c9ba96?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/4f0a67c9ba96</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Sat, 21 Feb 2026 02:17:36 GMT</pubDate>
            <atom:updated>2026-03-08T04:47:16.154Z</atom:updated>
            <content:encoded><![CDATA[<p>Self-hosted AI agent with Kimi Claw bridge and DeepSeek LLM</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/961/1*oYJqyY2CDj56z3selU2i2w.png" /></figure><h4><strong>Flow summary:</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/995/1*ebhzRGQ7EJfHLb5k0_ZZ8A.png" /></figure><h4>Step 1 — Deploy the Azure VM</h4><p>Create a resource group, VM, and lock down the NSG so only your IP can reach it.</p><pre># Variables — change these to match your environment<br>RESOURCE_GROUP=&quot;openclaw-rg&quot;<br>VM_NAME=&quot;openclawvm&quot;<br>LOCATION=&quot;uksouth&quot;<br>MY_IP=&quot;$(curl -s ifconfig.me)/32&quot;<br><br># Create resource group<br>az group create --name $RESOURCE_GROUP --location $LOCATION<br><br># Create VM (Ubuntu 24.04, Standard_D4s_v6)<br>az vm create --resource-group $RESOURCE_GROUP --name $VM_NAME --image Canonical:ubuntu-24_04-lts:server:latest --size Standard_D4s_v6 --admin-username azureadmin --generate-ssh-keys --public-ip-sku Standard<br><br># Get the NSG name (auto-created with the VM)<br>NSG_NAME=$(az network nsg list --resource-group $RESOURCE_GROUP --query &quot;[0].name&quot; -o tsv)<br><br># Allow SSH (port 22) from your IP only<br>az network nsg rule create --resource-group $RESOURCE_GROUP --nsg-name $NSG_NAME --name allow-ssh-myip --priority 300 --source-address-prefixes $MY_IP --destination-port-ranges 22 --protocol Tcp --access Allow<br><br># Allow All TCP ports from your IP only<br>az network nsg rule create --resource-group $RESOURCE_GROUP --nsg-name $NSG_NAME --name allow-https-myip --priority 310 --source-address-prefixes $MY_IP --destination-port-ranges * --protocol Tcp --access Allow<br><br># Get the VM&#39;s public IP<br>VM_IP=$(az vm show --resource-group $RESOURCE_GROUP --name $VM_NAME  --show-details --query publicIps -o tsv)<br><br>echo &quot;VM IP: $VM_IP&quot;</pre><h4>Step 2 — SSH into the VM and Install Prerequisites</h4><pre>ssh azureadmin@$VM_IP<br><br>sudo su -<br><br>apt update -y<br><br># Install Git and Curl<br>apt install -y git curl tmux<br><br># Install Node.js 22+<br>apt install -y nodejs<br>node --version</pre><h4>Step 3 — Install OpenClaw</h4><p>Run the official installer from <a href="https://openclaw.ai/">https://openclaw.ai/</a>:</p><pre>curl -fsSL https://openclaw.ai/install.sh | bash</pre><p>The installer will upgrade Node.js to v22 if needed, install build tools, install the openclaw npm package, and launch the onboarding wizard.</p><p><strong>During the onboarding wizard, choose:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/966/1*tW9oba0QZZfolCFl1omBvg.png" /></figure><p>The wizard will report that systemd user services are unavailable — this is expected on a root-only VM. We’ll run the gateway manually instead.</p><h4>Step 4 — Install and Configure nginx (HTTPS Reverse Proxy)</h4><p>OpenClaw’s Control UI requires a secure context (HTTPS or localhost). Since you’re accessing the VM over a public IP, you need nginx with a self-signed TLS certificate.</p><h4>Install nginx and generate a certificate</h4><pre>apt install -y nginx<br><br>openssl req -x509 -nodes -days 365 -newkey rsa:2048 \<br>  -keyout /etc/ssl/private/openclaw.key \<br>  -out /etc/ssl/certs/openclaw.crt \<br>  -subj &#39;/CN=openclaw&#39;</pre><h4>Configure the reverse proxy</h4><pre>cat &gt; /etc/nginx/sites-available/openclaw &lt;&lt; &#39;EOF&#39;<br>server {<br>    listen 443 ssl;<br>    server_name _;<br>    ssl_certificate /etc/ssl/certs/openclaw.crt;<br>    ssl_certificate_key /etc/ssl/private/openclaw.key;<br>    location / {<br>        proxy_pass http://127.0.0.1:18789;<br>        proxy_http_version 1.1;<br>        proxy_set_header Upgrade $http_upgrade;<br>        proxy_set_header Connection &quot;upgrade&quot;;<br>        proxy_set_header Host $host;<br>    }<br>}<br>EOF<br>ln -sf /etc/nginx/sites-available/openclaw /etc/nginx/sites-enabled/<br>rm -f /etc/nginx/sites-enabled/default<br>nginx -t &amp;&amp; systemctl restart nginx</pre><h4>Verify Nginx status</h4><pre># Verify nginx<br>systemctl start nginx<br>systemctl enable nginx<br>systemctl status nginx</pre><h4>Step 5 — Configure the Gateway (Loopback Bind)</h4><p>Edit the OpenClaw config to bind the gateway to loopback. This is required because the CLI refuses plaintext WebSocket connections to non-loopback addresses.</p><pre>vim ~/.openclaw/openclaw.json</pre><p>Set the gateway bind to loopback:</p><pre>{<br>  &quot;gateway&quot;: {<br>    &quot;bind&quot;: &quot;loopback&quot;<br>  }<br>}</pre><h4>Step 6 — Start the Gateway</h4><p>Since systemd user services are unavailable, run the gateway in a detached tmux session:</p><pre>tmux new -d -s openclaw &#39;openclaw gateway run&#39;</pre><h4>Step 7 — Pair Your Browser</h4><p>Get the token from the config:</p><pre>grep &#39;&quot;token&quot;&#39; ~/.openclaw/openclaw.json<br><br>## OR from the output of this command<br>openclaw dashboard</pre><p>Open the Control UI in your browser:</p><pre>https://&lt;VM-PUBLIC-IP&gt;/#token=&lt;your-gateway-token&gt;</pre><p>Your browser will warn about the self-signed certificate — click <strong>Advanced → Proceed</strong>.</p><p>The browser will show “disconnected — pairing required”. Approve it from the CLI:</p><pre>openclaw devices list                       # note the Request ID<br>openclaw devices approve &lt;request-id&gt;       # approve immediately</pre><blockquote><strong><em>Tip:</em></strong><em> Pairing request IDs expire quickly. If </em><em>approve fails with &quot;unknown requestId&quot;, re-run </em><em>devices list and approve the new ID right away.</em></blockquote><h4>Step 8 — Install Kimi Claw</h4><p>Go to <a href="https://www.kimi.com/">https://www.kimi.com/</a> ans sign in with your credentials.</p><p>On the left side, select <strong>Kimi Claw (Beta)</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/285/1*8Fz47RI5_rE95zpGPuBaNQ.png" /></figure><p>Then click on the link to “Link an existing OpenClaw setup”. This will give you the command and bot token to use to install the bridge plug-in.</p><pre>bash &lt;(curl -fsSL https://cdn.kimi.com/kimi-claw/install.sh) \<br>  --bot-token &lt;YOUR-KIMI-BOT-TOKEN&gt;</pre><p>The installer will download the plugin, register it in the OpenClaw config, and restart the gateway automatically.</p><p>Verify the bridge is connected:</p><pre>openclaw devices list<br><br># If you see &quot;Pending&quot;, review and approve the request<br>openclaw devices approve &lt;request-id&gt;</pre><p>You should see kimi-bridge-connector in the Paired devices list with an operator role.</p><h4>Step 9 — Configure the LLM Provider (DeepSeek)</h4><p>By default, OpenClaw expects Anthropic credentials. To use DeepSeek instead, edit the config via the Control UI’s <strong>Raw editor</strong> (Config → Raw) or directly:</p><pre>vim ~/.openclaw/openclaw.json</pre><p>Add the DeepSeek provider under the models key:</p><pre>{<br>  &quot;models&quot;: {<br>    &quot;mode&quot;: &quot;merge&quot;,<br>    &quot;providers&quot;: {<br>      &quot;deepseek&quot;: {<br>        &quot;baseUrl&quot;: &quot;https://api.deepseek.com&quot;,<br>        &quot;apiKey&quot;: &quot;&lt;YOUR-DEEPSEEK-API-KEY&gt;&quot;,<br>        &quot;api&quot;: &quot;openai-completions&quot;,<br>        &quot;models&quot;: [<br>          {<br>            &quot;id&quot;: &quot;deepseek-chat&quot;,<br>            &quot;name&quot;: &quot;DeepSeek Chat&quot;,<br>            &quot;reasoning&quot;: false,<br>            &quot;input&quot;: [&quot;text&quot;],<br>            &quot;cost&quot;: { &quot;input&quot;: 0, &quot;output&quot;: 0, &quot;cacheRead&quot;: 0, &quot;cacheWrite&quot;: 0 },<br>            &quot;contextWindow&quot;: 128000,<br>            &quot;maxTokens&quot;: 8192<br>          },<br>          {<br>            &quot;id&quot;: &quot;deepseek-reasoner&quot;,<br>            &quot;name&quot;: &quot;DeepSeek Reasoner&quot;,<br>            &quot;reasoning&quot;: true,<br>            &quot;input&quot;: [&quot;text&quot;],<br>            &quot;cost&quot;: { &quot;input&quot;: 0, &quot;output&quot;: 0, &quot;cacheRead&quot;: 0, &quot;cacheWrite&quot;: 0 },<br>            &quot;contextWindow&quot;: 128000,<br>            &quot;maxTokens&quot;: 65536<br>          }<br>        ]<br>      }<br>    }<br>  },<br>  &quot;agents&quot;: {<br>    &quot;defaults&quot;: {<br>      &quot;model&quot;: {<br>        &quot;primary&quot;: &quot;deepseek/deepseek-chat&quot;,<br>        &quot;fallbacks&quot;: []<br>      },<br>      &quot;models&quot;: {<br>        &quot;deepseek/deepseek-chat&quot;: {},<br>        &quot;deepseek/deepseek-reasoner&quot;: {}<br>      }<br>    }<br>  }<br>}</pre><p>Then restart the gateway:</p><pre>tmux kill-session -t openclaw 2&gt;/dev/null<br>tmux new -d -s openclaw &#39;openclaw gateway run&#39;</pre><h4>Verification</h4><p>Once everything is configured, you should be able to:</p><ol><li><strong>Open the Control UI</strong> at https://&lt;VM-PUBLIC-IP&gt;/#token=&lt;token&gt; and chat with the agent via the browser.</li><li><strong>Chat via Kimi Claw</strong> at <a href="https://kimi.com/">kimi.com</a> — messages route through the bridge to your self-hosted gateway, which calls DeepSeek for responses.</li></ol><p>Confirm the bridge is healthy:</p><pre>openclaw devices list</pre><p>You should see three paired devices: your browser, the kimi-bridge-connector, and the CLI session.</p><h4>Quick Command Reference</h4><pre># Useful tmux commands:<br>tmux attach -t openclaw          # view live logs (Ctrl+B then D to detach)<br>tmux kill-session -t openclaw    # stop the gateway<br><br># Restart pattern (after any config change):<br>tmux kill-session -t openclaw 2&gt;/dev/null<br><br># Start gateway<br>tmux new -d -s openclaw &#39;openclaw gateway run&#39;<br><br># Stop gateway<br>tmux kill-session -t openclaw<br><br># Restart gateway (after config changes)<br>tmux kill-session -t openclaw 2&gt;/dev/null<br>tmux new -d -s openclaw &#39;openclaw gateway run&#39;<br><br># View live logs<br>tmux attach -t openclaw           # Ctrl+B then D to detach<br><br># List / approve devices<br>openclaw devices list<br>openclaw devices approve &lt;request-id&gt;<br><br># Obtain the gateway token<br>cat ~/.openclaw/openclaw.json | grep -A2 &#39;&quot;auth&quot;&#39;<br><br># Obtain the Kimi bridge token<br>cat ~/.openclaw/openclaw.json | grep -A2 &#39;&quot;bridge&quot;&#39;<br><br># Check health<br>openclaw doctor<br><br># Follow log file<br>openclaw logs --follow</pre><h4>Security Notes</h4><ul><li><strong>NSG rules</strong> should be scoped to your IP only. Remove any broad “allow any IP” rules.</li><li><strong>Gateway token</strong> acts as a root password — keep it secret.</li><li><strong>Self-signed certificate</strong> triggers browser warnings. For for secure validation, use Let’s Encrypt with a real domain.</li><li><strong>Credential files</strong> should be locked down: chmod 600 ~/.openclaw/credentials/*</li><li><strong>Consent mode</strong> can be enabled in config (exec.ask: &quot;on&quot;) to require approval before the agent executes commands.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4f0a67c9ba96" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My thoughts on LLMs as a tech writer (Part 2)]]></title>
            <link>https://davidokeyode.medium.com/my-thoughts-on-llms-as-a-tech-writer-part-2-eb80b39ca4d1?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/eb80b39ca4d1</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Thu, 19 Feb 2026 04:05:26 GMT</pubDate>
            <atom:updated>2026-02-22T15:45:33.528Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*R-MA_A7tkXZa6XLLcTfJ2Q.png" /></figure><p>In <a href="https://medium.com/@davidokeyode/the-slop-problem-will-llms-replace-thinking-1f2313c73533">Part 1</a>, I reflected on how LLMs can be both useful and harmful for writers . They are useful when they sharpen our thinking, harmful when they replace it. The danger is they can make it tempting to <em>skip thinking altogether</em>. Research shows that when people use AI assistance, they reduce their engagement and effort. This means that they <a href="https://www.nature.com/articles/s41598-025-98385-2">offload their thinking</a> to <a href="https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf">AI altogether</a>.</p><p>I also compared the flood of low quality AI-generated content (slop), to the spam epidemic that followed the rise of email communication. I asked whether we’d eventually need filters for AI slop the way we needed spam filters for our inboxes. As it turns out, we’re starting to see attempts to allow users to filter out these low quality content. Platforms like Pinterest now allows users to reduce AI-generated images in their feeds. TikTok introduced measures in late 2025 to curb synthetic media. Spotify is labeling and filtering out AI-generated music. And while browser extensions to filter text-based slop haven’t seen mass adoption yet, their very existence tells us something about the dissatisfaction out there.</p><p>We’re in an era of homogenized voices. More “information” is sounding the same because more people are outsourcing their expressions to the same AI models. Distinctiveness is gradually collapsing! The genuine voices will stand out. The writers who can may use AI tools to fine-tune their expressions while maintaining their unique human texture, will be uncommon.</p><p>This brings me to a skill I think writers now need to deliberately cultivate: the ability to write in a way that doesn’t sound like an LLM, even when using one. Use the tool to refine your thinking, but protect what makes your writing recognizable.</p><p>Mark Dingemanse’s point from <a href="https://medium.com/@davidokeyode/the-slop-problem-will-llms-replace-thinking-1f2313c73533">Part 1</a> is an important one: “Writing is thinking.” I’d extend it further by saying: writing is thinking in <strong><em>your particular way</em></strong>, and expressing it in <strong><em>your particular voice</em></strong>. Don’t let any tool take that away from you. Don’t let the pressure to “force multiply,” or to be endlessly productive push you into sounding like everyone else.</p><p>My writing is an expression of who I am. I can’t outsource who I am to a machine. I have to stay true to who I am.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eb80b39ca4d1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hands-On AI Security Labs — MCP Breach-to-Fix Labs - Lab 01: Asana Multi-Tenant Authorization…]]></title>
            <link>https://davidokeyode.medium.com/hands-on-ai-security-labs-mcp-breach-to-fix-labs-lab-01-asana-multi-tenant-authorization-c363c2295dc5?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/c363c2295dc5</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Thu, 12 Feb 2026 16:25:26 GMT</pubDate>
            <atom:updated>2026-02-12T16:30:05.841Z</atom:updated>
            <content:encoded><![CDATA[<h3>Hands-On AI Security Labs — MCP Breach-to-Fix Labs - Lab 01: Asana Multi-Tenant Authorization Bypass</h3><p>This lab is based on the MCP Breach-to-Fix Labs repository by <a href="https://github.com/PawelKozy">Pawel Kozy</a>. Huge thanks to him for building and sharing these practical MCP security challenges: <a href="https://github.com/PawelKozy/mcp-breach-to-fix-labs">https://github.com/PawelKozy/mcp-breach-to-fix-labs</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8CdG77WU2-GRlnHzAhvBJA.png" /></figure><p><strong>Real-World Incident</strong> — In June 2025, Asana discovered a serious bug in their MCP server that had been live since May 1st 2025. The vulnerability allowed data belonging to one organization to be visible to users from other organizations. For nearly two weeks (June 5–17), Asana took their MCP integration completely offline while they investigated and patched the issue. Approximately 1,000 customers were potentially affected, including Fortune 500 companies. The incident reportedly cost Asana $7.5 million to remediate. More details: <a href="https://www.upguard.com/blog/asana-discloses-data-exposure-bug-in-mcp-server">https://www.upguard.com/blog/asana-discloses-data-exposure-bug-in-mcp-server</a></p><h3>Lab 01: Asana Multi-Tenant Authorization Bypass</h3><p>This guide walks you through setting up and exploiting a real-world MCP vulnerability pattern. The objective is to demonstrate this common multi-tenant security failure where the server correctly verifies that an API key is valid (authentication), but never checks whether the requested project belongs to that tenant (authorization). This results in unauthorized cross-tenant access if exploited.</p><h4>PREREQUISITES</h4><p>Before starting, ensure you have:</p><ol><li><strong>Docker Desktop for Windows</strong> — <a href="https://www.docker.com/products/docker-desktop/">https://www.docker.com/products/docker-desktop/</a></li><li><strong>Git for Windows</strong> — <a href="https://git-scm.com/download/win">https://git-scm.com/download/win</a></li><li><strong>Cursor IDE</strong> — <a href="https://cursor.sh/">https://cursor.sh/</a></li></ol><h3>STEP 1: Clone the Repository</h3><p>Open PowerShell and run:</p><pre>mkdir &quot;$env:USERPROFILE\GitHubRepos&quot;<br>cd &quot;$env:USERPROFILE\GitHubRepos&quot;<br>git clone https://github.com/PawelKozy/mcp-breach-to-fix-labs.git<br>cd mcp-breach-to-fix-labs</pre><h3>STEP 2: Start Docker Desktop</h3><ol><li>Press the Windows key and type “Docker Desktop”</li><li>Open Docker Desktop</li><li>Wait for it to fully start (whale icon in system tray shows green/running)</li></ol><p>Verify it’s running:</p><pre>docker info</pre><h3>STEP 3: Start the Vulnerable Server</h3><p>Run this command:</p><pre>docker compose up challenge-01-vulnerable</pre><p>Wait for the build to complete. You should see:</p><pre>challenge-01-vulnerable-1  | INFO: Uvicorn running on http://0.0.0.0:8000</pre><p>Keep this PowerShell window open. The server runs on port 8001.</p><h3>STEP 4: Configure the MCP Client (Cursor)</h3><h4>4.1 Create the configuration file</h4><p>Open a NEW PowerShell window and run:</p><pre># Cursor Configuration<br>mkdir -Force &quot;$env:USERPROFILE\.cursor&quot;<br>notepad &quot;$env:USERPROFILE\.cursor\mcp.json&quot;</pre><h4>4.2 Paste this exact content into Notepad</h4><pre>{<br>  &quot;mcpServers&quot;: {<br>    &quot;challenge-01-vulnerable&quot;: {<br>      &quot;url&quot;: &quot;http://localhost:8001/mcp/stream&quot;<br>    }<br>  }<br>}</pre><h4>4.3 Save and close</h4><p>Press Ctrl+S to save, then close Notepad.</p><h3>4.4 Restart the MCP client completely</h3><ol><li>Close and reopen Cursor.</li><li>When prompted to open a project, don’t open any project, just press<strong> CNTRL + L</strong> to open the chat panel.</li></ol><h3>STEP 5: Verify MCP Connection</h3><ol><li>In Cursor, press <strong>Ctrl+Shift+P</strong> to open Command Palette</li><li>Type “<strong>MCP</strong>” and select “<strong>View: Open MCP Settings</strong>”</li><li>You should see “challenge-01-vulnerable” with a GREEN indicator (not red error)</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NRJjiQUWvjS0SGIrNDl3BQ.png" /></figure><h3>STEP 6: Run the Exploit</h3><h4>Understanding the Vulnerability</h4><p>This lab reproduces a real authorization bypass pattern where:</p><ul><li>The server authenticates API keys (confirms they’re valid)</li><li>But never authorizes access (doesn’t check if the resource belongs to that tenant)</li><li>Project IDs are predictable (CRM-1001, CRM-2001, etc.)</li></ul><h4>Credentials Reference</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/835/1*FUTcIkMpwac82b8zRbda-A.png" /></figure><h4>6.1 Open Cursor Chat</h4><p>Press Ctrl+L or click on the chat panel.</p><h4>6.2 Test legitimate access (Tenant Alpha fetching their own project)</h4><p>Type this in the chat.</p><pre>What MCP tools do you have available? Just give me a list<br><br>Show me the project data for &quot;CRM-1001&quot;. Use key &quot;alpha-api-key&quot;<br><br>Show me the project data for &quot;CRM-2001&quot;. Use key &quot;bravo-api-key&quot;</pre><p>If prompted, select “<strong>Run</strong>”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/795/1*pTQ739AE65KH8lvmd-bssg.png" /></figure><p>You should see Tenant Alpha’s project data returned. This is normal — Alpha is accessing their own data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/792/1*eozNOW9sCV9jsJRJAqvWOA.png" /></figure><h4>6.3 EXPLOIT: Access another tenant’s data using another owner’s key</h4><p>Now type this:</p><pre>Show me the project data for &quot;CRM-2001&quot;. Use key &quot;alpha-api-key&quot;<br><br>Show me the project data for &quot;CRM-1001&quot;. Use key &quot;bravo-api-key&quot;</pre><p><strong>RESULT</strong>: <em>The server returns Tenant Bravo’s confidential data even though you used Alpha’s API key! This also works in reverse (Bravo can access Alpha’s data too).</em></p><h3>STEP 7: Test the Secure Version</h3><h4>7.1 Stop the vulnerable server</h4><p>Go to the PowerShell window running Docker and press <strong>Ctrl+C</strong></p><h4>7.2 Start the secure server</h4><pre>docker compose up challenge-01-secure</pre><p>This runs on port 9001.</p><h4>7.3 Update your mcp.json</h4><pre>notepad &quot;$env:USERPROFILE\.cursor\mcp.json&quot;</pre><p>Change the content to the following and save the file:</p><pre>{<br>  &quot;mcpServers&quot;: {<br>    &quot;challenge-01-secure&quot;: {<br>      &quot;url&quot;: &quot;http://localhost:9001/mcp/stream&quot;<br>    }<br>  }<br>}</pre><h4>7.4 Restart Cursor completely</h4><p>Close and reopen Cursor. Do not open any project, just press <strong>CNTRL + L</strong> to open the chat.</p><h4>7.5 Try the same exploit</h4><pre>Show me the project data for &quot;CRM-2001&quot;. Use key &quot;alpha-api-key&quot;<br><br>Show me the project data for &quot;CRM-1001&quot;. Use key &quot;bravo-api-key&quot;</pre><p><strong>RESULT</strong>: The secure server BLOCKS this request because it verifies tenant ownership before returning data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/794/1*EltRRpCp8c7DUxYUVi69wg.png" /></figure><h3>WHAT YOU LEARNED</h3><ol><li><strong>Authentication vs Authorization</strong>: The vulnerable server authenticates (verifies the API key is valid) but doesn’t authorize (verify the user should access that specific resource).</li><li><strong>Confused Deputy Attack</strong>: The MCP proxy uses a single high-privilege service token for all backend requests, trusting the caller without verification.</li><li><strong>Predictable IDs</strong>: Sequential IDs (CRM-1001, CRM-2001) make enumeration attacks trivial.</li><li><strong>The Fix</strong>: The secure version adds tenant-scoped authorization checks before returning any data.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c363c2295dc5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI for the Cloud Practitioner — Part III— Creating an n8n AI Agent with Azure Storage Integration]]></title>
            <link>https://davidokeyode.medium.com/ai-for-the-cloud-practitioner-part-iii-creating-an-n8n-ai-agent-with-azure-storage-integration-8ace69f9b308?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/8ace69f9b308</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Wed, 13 Aug 2025 18:59:21 GMT</pubDate>
            <atom:updated>2025-08-13T18:59:21.554Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI for the Cloud Practitioner — Part III— Creating an n8n AI Agent with Azure Storage Integration</h3><p><a href="https://n8n.io/">N8n</a> provides a platform to create AI agents using visual tools (no code). It has various ways to integrate with Azure platform and services. In this blog post, I’ll walk you through how you can create an AI agent in n8n that integrates with a NoSQL database in Azure Table Storage. This post is aimed at beginners.</p><h3>What You’ll Build</h3><p>The AI assistant will:</p><ul><li>Search through student records stored in Azure Table Storage</li><li>Answer questions about student information</li><li>Update student records through natural language commands</li><li>Remember conversation context</li></ul><p>We’ll walk through this process in four parts:</p><ol><li>Setting up n8n locally</li><li>Creating test data in Azure Table Storage</li><li>Configuring OpenAI integration</li><li>Building the complete workflow with AI agent and tools</li></ol><p>This is the final result that we will have at the end of this guide:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/770/1*BdDTx4hgdBgVEgc7f38BCw.png" /></figure><h3>Part 1: Setting Up n8n locally</h3><p>First, install the <a href="https://docs.n8n.io/hosting/community-edition-features/">community edition of <strong>n8n</strong></a><strong> </strong>on your PC. Open PowerShell as an administrator and run these commands. The commands will install <strong>Nodejs </strong>and <strong>NPM </strong>using Chocolatey and then run <strong>n8n </strong>using <strong>npx</strong>.</p><pre># Install Chocolatey<br>Set-ExecutionPolicy Bypass -Scope Process -Force; [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; iex ((New-Object System.Net.WebClient).DownloadString(&#39;https://community.chocolatey.org/install.ps1&#39;))<br><br># Install Node.js and npm<br>choco install nodejs.install -y<br><br># Refresh and confirm<br>Import-Module $env:ChocolateyInstall\helpers\chocolateyProfile.psm1<br><br>refreshenv<br><br>node --version<br>npm --version<br>npx --version<br><br>## Start n8n without installing<br>npx n8n<br><br>Open your browser and go to http://localhost:5678</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/558/1*D2gE46MBQHnoT4aR2VGtxg.png" /></figure><h3>Part 2: Creating a storage account with test data in Azure</h3><p>You’ll need an Azure Storage account with sample data so we can test. You can do this in Cloud Shell (Bash).</p><h4>Step 1: Create a storage account with a table called “<strong>studentsDB</strong>”. Add sample data to it.</h4><pre>random=$RANDOM<br>RG=&quot;ai-agent-rg&quot;<br>LOC=&quot;uksouth&quot;<br>SA=&quot;storeacct$random&quot;<br><br>az group create --name $SA --resource-group $RG --location $LOC<br><br>az storage account create --name $SA --resource-group $RG --location $LOC --sku Standard_LRS --kind StorageV2<br><br>SA_KEY=$(az storage account keys list --resource-group $RG --account-name $SA --query &quot;[0].value&quot; -o tsv)<br><br>az storage table create --name studentsDB --account-name $SA --account-key $SA_KEY<br><br>export PYTHONWARNINGS=&quot;ignore::UserWarning&quot;<br><br>CSV_URL=&quot;https://raw.githubusercontent.com/davidokeyode/azure-offensive/refs/heads/master/students_sample.csv&quot;<br>TMP_CSV=$(mktemp)<br><br>curl -sSL &quot;$CSV_URL&quot; -o &quot;$TMP_CSV&quot;<br><br>tail -n +2 &quot;$TMP_CSV&quot; | while IFS=&#39;,&#39; read -r pk rk fn ln major year gpa lastrev advisor; do<br> az storage entity insert --table-name studentsDB --account-name &quot;$SA&quot; --account-key &quot;$SA_KEY&quot; --entity PartitionKey=&quot;$pk&quot; RowKey=&quot;$rk&quot; FirstName=&quot;$fn&quot; LastName=&quot;$ln&quot; Major=&quot;$major&quot; EnrollmentYear=&quot;$year&quot; GPA=&quot;$gpa&quot; LastReview=&quot;$lastrev&quot; StudentAdvisor=&quot;$advisor&quot; --only-show-errors<br>done<br><br>rm -f &quot;$TMP_CSV&quot;<br>unset PYTHONWARNINGS</pre><h4>Step 2: Generate a SAS URL.</h4><p>Make a note of the SAS_URL output.</p><pre>EXPIRY=$(date -u -d &quot;+7 days&quot; &#39;+%Y-%m-%dT%H:%MZ&#39;)<br><br>TABLE_SAS=$(az storage table generate-sas --name studentsDB --account-name &quot;$SA&quot; --account-key  &quot;$SA_KEY&quot; --permissions raud --expiry &quot;$EXPIRY&quot; --https-only -o tsv)<br><br>SAS_URL=&quot;https://$SA.table.core.windows.net/studentsDB?$TABLE_SAS&quot;<br><br>echo &quot;Make a note of this SAS URL: $SAS_URL&quot;</pre><h3>Part 3: Creating OpenAI API key and add to n8n</h3><h4>Step 1: Create API key in Open AI</h4><ul><li><a href="https://platform.openai.com/api-keys">https://platform.openai.com/api-keys</a> → Create new secret key</li><li><strong>Name</strong>: n8n-local</li><li><strong>Project</strong>: Default project</li><li>Click <strong>Create secret key</strong></li><li>Make a note of the secret key as you will need it in the next step</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/726/1*LQ91-VtzZD2TuVtVaNghIQ.png" /></figure><h4>Step 2: Add the OpenAI credential in n8n</h4><p><strong>n8n </strong>→ Overview → Create Credential</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-SZ9gChN0Y4Ii-wpy7j_Ug.png" /></figure><p><strong>Search for app</strong>: OpenAi<br>Click <strong>Continue</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/449/1*TYxzOAOBzd0oPTPFbf8YqA.png" /></figure><p>In the <strong>Connection </strong>tab<br><strong>API Key</strong>: Paste the API key from the previous step<br>Click <strong>Save</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/702/1*HtAi29pKegSHyDxWZRuy0g.png" /></figure><h3>Part 4: Create n8n workflow</h3><h4>Step 1: Create Workflow</h4><p>In<strong> n8n </strong>→ Overview → Create Workflow → Add first step → On chat message</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/557/1*7xd24a_G3jQnSuQsL9Witg.png" /></figure><p>On the “<strong>When a chat message is received</strong>” menu, configure the following:</p><ul><li><strong>Make Chat Publicly Available:</strong> Enabled</li><li><strong>Mode:</strong> Hosted chat</li><li><strong>Authentication:</strong> None</li><li><strong>Initial message:</strong> “Hello! I’m a Student Advisor AI assistant. How can I assist you today?”</li><li>Leave other fields at default value</li><li>Click <strong>Back to Canvas</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/384/1*ZRW0pxBl5_lvjYu2gLh7vw.png" /></figure><p>The canvas should be similar to the below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/440/1*DUpmUgr59pjwxemu1QCWyg.png" /></figure><h4>Step 2: Add an AI agent with a model</h4><p>Click ADD NEXT STEP → AI → AI Agent</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/684/1*j7hyQOAjHD95zvjsJpLfYg.png" /></figure><p>In the <strong>AI Agent</strong> menu, leave all settings as default. Click <strong>Back to canvas</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/543/1*Lw1-yBr0WPyAtHm_OQqLjA.png" /></figure><p>Back in the canvas, click to add a <strong>Chat Model</strong>, select<strong> OpenAI Chat Model</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KYmp2k2I4HdMqD-CiWlReQ.png" /></figure><p>In the OpenAI Chat Model, configure the following:</p><ul><li><strong>Credential</strong>: OpenAi account</li><li><strong>Model</strong>: gpt-4o</li><li>Click <strong>Back to canvas</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/609/1*x29cRHKDsekHtAw4_LbSeQ.png" /></figure><p>The canvas should be similar to the below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/626/1*MB1AvB9soUP9cH-8O_ahAQ.png" /></figure><h4>Step 3: Add MEMORY to the AI agent</h4><p>Back in the canvas, click to add <strong>Memory</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Do3WFY-gK5iCTE64cZXmmg.png" /></figure><p>In the <strong>Simple Memory</strong> menu, configure the following:</p><ul><li><strong>Context Windows Length</strong>: 50</li><li>Leave all settings as default.</li><li>Click <strong>Back to canvas</strong>.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/724/1*t0WYHGY6Ztfb8FcknQgtTA.png" /></figure><p>The canvas should be similar to the below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/639/1*RkqsO5DcUzWknlYNmmVHkQ.png" /></figure><h4>Step 4: Add the first TOOL to the AI agent</h4><p>Back in the canvas, click to add a <strong>Tool</strong>, select <strong>HTTP Request Tool</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/978/1*E6x89ubvmIszfcS9l7haOQ.png" /></figure><p>In the <strong>HTTP Request </strong>menu, configure the following:</p><ul><li><strong>Description</strong>: <em>Searches the studentsDB table in Azure Storage. The entities have the following properties: PartitionKey, RowKey, FirstName, LastName, Major, EnrollmentYear, GPA, LastReview, StudentAdvisor.</em></li><li><strong>Method</strong>: GET</li><li><strong>URL</strong>: Paste the SAS URL from the earlier step</li><li><strong>Authentication</strong>: None</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/601/1*0LIRIucdVhyFF38kKBrovQ.png" /></figure><ul><li><strong>Send Query Parameters</strong>: Enabled</li></ul><p>— <strong>Name </strong>= $filter ; <strong>Value (Expression) </strong>= {{ $fromAI(&#39;filter_condition&#39;, &#39;&#39; , &#39;string&#39;) }}</p><p>— <strong>Name </strong>= $top ; <strong>Value (Expression) </strong>= {{ $fromAI(&#39;record_limit&#39;, &#39;100&#39; , &#39;string&#39;) }}</p><p>— <strong>Name </strong>= $select ; <strong>Value (Expression) </strong>= {{ $fromAI(&#39;properties_to_select&#39;, &#39;PartitionKey,RowKey&#39;, &#39;string&#39;) }}</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*_fr-bsnarVlCnKz7Xaa4MQ.png" /></figure><ul><li><strong>Send Headers</strong>: Enabled</li></ul><p>— <strong>Name </strong>= $Accept ; <strong>Value (Fixed) </strong>= application/json;odata=nometadata</p><p>— <strong>Name </strong>= $x-ms-version ; <strong>Value (Fixed) </strong>= 2020-08-04</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/586/1*zwEkqQdwgNhgHThq-VpbZA.png" /></figure><ul><li>Leave other settings as default.</li><li>Click <strong>Back to canvas</strong>.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/724/1*t0WYHGY6Ztfb8FcknQgtTA.png" /></figure><p>Rename the node to <strong>Search StudentsDB</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/852/1*cIgeX_4aBzecf3LqaG961A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/413/1*6uXS7HP3yGfeVaurk7cvTg.png" /></figure><p>The canvas should be similar to the below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/702/1*Djwml5Hjye0BHXyPZvd8Fw.png" /></figure><h4>Step 5: Add the second TOOL to the AI agent</h4><p>Back in the canvas, click to add a <strong>Tool</strong>, select <strong>HTTP Request Tool</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/978/1*E6x89ubvmIszfcS9l7haOQ.png" /></figure><p>In the <strong>HTTP Request </strong>menu, configure the following:</p><ul><li><strong>Description</strong>: <em>Updates or inserts an entity in the studentsDB table in Azure Storage: supply PartitionKey and RowKey to locate the row, then include any properties to change (FirstName, LastName, Major, EnrollmentYear, GPA, LastReview, StudentAdvisor); the node issues a Table REST “MERGE” so unchanged fields stay intact.</em></li><li><strong>Method</strong>: POST</li><li><strong>URL</strong>: Defined automatically by the model</li><li><strong>Add a description</strong>: Replace <strong>storeacctXXXXX</strong> with your actual storage account name and the value for <strong>SAS</strong> with your actual SAS value (the section in the <strong>SAS URL</strong> from <strong>?sv</strong>)</li></ul><pre>Return the full entity-URI for Azure Table Storage:<br>https://storeacctXXXXX.table.core.windows.net/studentsDB(PartitionKey=&#39;{PK}&#39;,RowKey=&#39;{RK}&#39;)?&lt;SAS&gt;<br> - replace {PK} with the URL-encoded PartitionKey and {RK} with the URL-encoded RowKey, then append the table-level SAS token unchanged. The partition key and row key of the record to change can be obtained from the &quot;Search StudentsDB&quot; tool.<br>SAS: ?sv=2024–11–04&amp;ss=t&amp;srt=sco&amp;sp=rwdlacu&amp;se=2025–07–19T21:27:57Z&amp;st=2025–07–05T13:27:57Z&amp;spr=https&amp;sig=3xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxD</pre><ul><li><strong>Authentication</strong>: None</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/670/1*m9uD2dvMvhysftuzOWHktw.png" /></figure><ul><li><strong>Send Query Parameters</strong>: Disabled</li><li><strong>Send Headers</strong>: Enabled</li></ul><p>— <strong>Name </strong>=Accept ; <strong>Value (Fixed) </strong>= application/json;odata=nometadata</p><p>— <strong>Name </strong>= Content-Type ; <strong>Value (Fixed) </strong>= application/json</p><p>— <strong>Name </strong>=X-HTTP-Method ; <strong>Value (Fixed) </strong>= MERGE</p><p>— <strong>Name </strong>=If-Match ; <strong>Value (Fixed) </strong>= *</p><p>— <strong>Name </strong>=x-ms-version ; <strong>Value (Fixed) </strong>= 2020-08-04</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/576/1*Y_Wld_LIMpfo7iGylaPXPA.png" /></figure><ul><li><strong>Send Body</strong>: Enabled</li></ul><p>— <strong>Body Content Type </strong>=JSON</p><p>— <strong>Specify Body </strong>=Using JSON</p><p>— <strong>JSON </strong>=Defined automatically by the model</p><p>— <strong>Description </strong>=</p><pre>JSON payload that Azure Table Storage accepts for a MERGE/UPDATE. Include PartitionKey and RowKey, then list only the properties to change. Numeric fields stay numbers; dates are ISO 8601 strings.<br>Sample:<br><br>{<br>  &quot;PartitionKey&quot;: &quot;CS&quot;,<br>  &quot;RowKey&quot;: &quot;S0001&quot;,<br>  &quot;FirstName&quot;: &quot;Amina&quot;,<br>  &quot;LastName&quot;: &quot;Ahmed&quot;,<br>  &quot;Major&quot;: &quot;Computer Science&quot;,<br>  &quot;EnrollmentYear&quot;: 2025,<br>  &quot;GPA&quot;: 3.85,<br>  &quot;LastReview&quot;: &quot;2025-07-05&quot;,<br>  &quot;StudentAdvisor&quot;: &quot;Dr. Adebayo&quot;<br>}<br><br>The node adds X-HTTP-Method: MERGE, If-Match: *, and Content-Type: application/json; Azure returns 204 No Content on success.</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/577/1*9-rrM7fUTMkAidGN-z4d6w.png" /></figure><ul><li>Click <strong>Back to canvas</strong>.</li></ul><p>Rename the node to <strong>Update StudentsDB</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/892/1*YyzPGyJhkkb_ywp2N7hfdA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/415/1*N6iPIIOQnJy9Za_f6B0uVg.png" /></figure><p>Click <strong>Save</strong>. The canvas should be similar to the below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/864/1*n0U02oxSOyCHXzrfRb7tGQ.png" /></figure><h4>Step 6: Test the workflow and AI agent</h4><p>In the lower left section of the canvas, try out a few prompts to retrieve and update information from the Tables database.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jy8Zw7uYFgKshMVyFFCFXg.png" /></figure><pre>Get the list of computer science majors names and their last review dates<br><br>Who is Liam Iyer&#39;s student advisor?<br><br>Update the student advisor of Liam Iyer to Dr. David Okeyode</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/819/1*MXHF5iq3Fkwr6QtH7kONTQ.png" /></figure><p>Congratulations! You have reached the end of this post. Don’t forget to cleanup resources that were created (especially the Azureresources).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8ace69f9b308" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI for the Cloud Practitioner — Part II— Create an AI Agent with an OpenAPI Action]]></title>
            <link>https://davidokeyode.medium.com/ai-for-the-cloud-practitioner-part-ii-create-an-ai-agent-with-an-openapi-action-9bf76dadd2c4?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/9bf76dadd2c4</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Sun, 27 Jul 2025 14:48:20 GMT</pubDate>
            <atom:updated>2025-07-27T14:48:20.940Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI for the Cloud Practitioner — Part II— Create an AI Agent with an OpenAPI Action</h3><p>This post walks you through building an AI agent that talks <strong>directly</strong> to the Azure DevOps REST API (no <a href="https://davidokeyode.medium.com/ai-for-the-cloud-practitioner-part-i-create-an-ai-foundry-agent-with-a-logic-app-action-cbbd4a47c37d">Logic App in the middle</a>) to create work items. This approach reduces platform cost and latency but requires advanced knowledge because you must understand the target API and authentication.</p><p>If you prefer to watch a demo, here’s a recorded meetup session: <a href="https://www.youtube.com/watch?v=udRCNE-w3c8&amp;ab_channel=DavidOkeyode">https://www.youtube.com/watch?v=udRCNE-w3c8&amp;ab_channel=DavidOkeyode</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rV193WBhYaX09OdqwtmR5w.png" /></figure><h3>Pre‑Requisites</h3><p>Before creating your AI agent with OpenAPI actions, make sure you have:</p><ol><li>An Azure subscription with <strong>Azure AI Foundry</strong> access</li><li>An <strong>Azure DevOps</strong> organization and project</li><li>An <strong>Azure DevOps Personal Access Token (PAT)</strong> with the scopes you want to allow the agent (e.g., Work Items: Read &amp; Write)</li></ol><h3>What We’re Building</h3><p>We’ll create a two‑part solution:</p><ol><li><strong>An AI Foundry Agent</strong> (GPT‑4o) that collects/validates user inputs and calls a tool.</li><li><strong>An OpenAPI 3.0 tool</strong> that sends a POST request to Azure DevOps’ <strong>Work Items</strong> endpoint to create a <strong>Bug</strong>.</li></ol><h4><strong>Key components</strong></h4><ul><li><strong>AI Foundry Agent</strong> — natural‑language interface and conversation orchestration</li><li><strong>Azure DevOps REST API</strong> — the target system for work item creation</li><li><strong>OpenAPI 3.0 schema</strong> — formal description of the endpoint &amp; authentication</li><li><strong>API Connection</strong> — securely provides the PAT via an Authorization header</li></ul><h3>Step 1 — Create an Azure DevOps Personal Access Token (PAT)</h3><p>You’ll use this token to authenticate the API call that creates work items.</p><ol><li>Navigate to Azure DevOps (dev.azure.com) → <strong>Top-right avatar</strong> → <strong>Personal access tokens</strong> → <strong>+ New Token</strong></li></ol><ul><li><strong>Name:</strong> ado-ai-agent-workitem</li><li><strong>Organization:</strong> Select the org where the project lives</li><li><strong>Expiration:</strong> Choose a short, sensible duration (e.g., 30 or 90 days)</li><li><strong>Scopes (minimum required): Work Items: Read &amp; write</strong> ← <em>Required to create work items</em></li><li>Click <strong>Create</strong>, then <strong>copy the token value</strong> and store it securely (you won’t be able to view it again).</li></ul><h3>Step 2 — Generate a Base64‑encoded PAT</h3><p>You’ll pass the PAT in an Authorization: Basic … header (username is empty, PAT is the password). Create the header value:</p><pre># Replace with your actual PAT<br>PAT=&quot;your_personal_access_token_here&quot;<br><br># Build &quot;Basic &lt;base64(:PAT)&gt;&quot; value<br>AUTH=$(printf &#39;:%s&#39; &quot;$PAT&quot; | base64 -w0)<br>echo &quot;Basic $AUTH&quot;</pre><p>Make a note of the output. You’ll use it in the connection step below.</p><h3>Step 3 — Create the API Connection in AI Foundry</h3><ol><li>Navigate to<strong> AI Foundry Project</strong> → <strong>Management Center</strong> → <strong>Project</strong> → <strong>Connected resources</strong> → <strong>+ New Connection</strong> → <strong>Other resource types</strong> → <strong>Custom keys</strong></li></ol><p><strong>Authentication Configuration:</strong></p><ul><li><strong>Authentication:</strong> Custom</li><li><strong>Add key/value pair</strong></li><li><strong>Name:</strong> Authorization</li><li><strong>Value:</strong> Basic &lt;base64_encoded_PAT&gt;( The output value from Step 2)</li><li><strong>IsSecret:</strong> ✔</li><li><strong>Connection name:</strong> azure_devops_conn_custom</li></ul><h3>Step 4 — AI Model Deployment</h3><ol><li>Navigate to <strong>AI Foundry Project</strong> → <strong>My assets</strong> → <strong>Models + endpoints</strong> → <strong>Model deployments</strong> → <strong>+ Deploy model</strong> → <strong>Deploy base model → </strong>Select <strong>gpt-4o</strong> and confirm deployment.</li></ol><h3>Step 5— Create the AI Agent</h3><ol><li>Navigate to<strong> AI Foundry Project</strong> → <strong>Build and customize</strong> → <strong>Agents + endpoints</strong> → <strong>My agents</strong> → <strong>+ New agent</strong></li></ol><p><strong>Agent Configuration:</strong></p><ul><li><strong>Agent name</strong>: ADO-Ticket-Creator-API</li><li><strong>Deployment</strong>: gpt-4o</li><li><strong>Agent Description</strong>: A friendly AI that greets each new thread, collects and validates a work item’s Title, Description, Priority (1–4), and Tags, then creates the Azure DevOps Bug work item and outputs the corresponding JSON payload.</li></ul><h3>Step 6 — Define the OpenAPI 3.0 Schema (Tool)</h3><p>Add OpenAPI action: <strong>Actions </strong>→ <strong>+ Add</strong> → <strong>OpenAPI 3.0 specified tool</strong></p><p><strong>Action Configuration:</strong></p><ul><li><strong>Name</strong>: azure_devops_work_item_create_api</li><li><strong>Description</strong>: Creates a new Bug work item in the specified Azure DevOps organization and project by sending a POST request to the Azure DevOps Work Items REST endpoint.</li><li><strong>Authentication method</strong>: Connection</li><li><strong>Choose connection</strong>: azure_devops_conn_custom</li></ul><h4>OpenAPI Schema</h4><ul><li><strong>Replace</strong> ORG and PROJECT with your Azure DevOps organization and project names in the schema below.</li></ul><pre>{<br>  &quot;openapi&quot;: &quot;3.0.1&quot;,<br>  &quot;info&quot;: {<br>    &quot;title&quot;: &quot;Create Azure DevOps Bug Work Item&quot;,<br>    &quot;description&quot;: &quot;Creates a new Bug work item in the Azure DevOps project of the ORG organization. The caller only needs to provide title, description, priority, and tags.&quot;,<br>    &quot;version&quot;: &quot;1.0.0&quot;<br>  },<br>  &quot;servers&quot;: [<br>    {<br>      &quot;url&quot;: &quot;https://dev.azure.com/ORG/PROJECT&quot;<br>    }<br>  ],<br>  &quot;components&quot;: {<br>    &quot;securitySchemes&quot;: {<br>      &quot;azureDevOpsPAT&quot;: {<br>        &quot;type&quot;: &quot;apiKey&quot;,<br>        &quot;in&quot;: &quot;header&quot;,<br>        &quot;name&quot;: &quot;Authorization&quot;<br>      }<br>    }<br>  },<br>  &quot;security&quot;: [<br>    {<br>      &quot;azureDevOpsPAT&quot;: []<br>    }<br>  ],<br>  &quot;paths&quot;: {<br>    &quot;/_apis/wit/workitems/$Bug&quot;: {<br>      &quot;post&quot;: {<br>        &quot;operationId&quot;: &quot;createBugWorkItem&quot;,<br>        &quot;summary&quot;: &quot;Create a Bug work item&quot;,<br>        &quot;parameters&quot;: [<br>          {<br>            &quot;name&quot;: &quot;api-version&quot;,<br>            &quot;in&quot;: &quot;query&quot;,<br>            &quot;required&quot;: true,<br>            &quot;schema&quot;: {<br>              &quot;type&quot;: &quot;string&quot;,<br>              &quot;default&quot;: &quot;7.1&quot;<br>            }<br>          }<br>        ],<br>        &quot;requestBody&quot;: {<br>          &quot;required&quot;: true,<br>          &quot;content&quot;: {<br>            &quot;application/json-patch+json&quot;: {<br>              &quot;schema&quot;: {<br>                &quot;type&quot;: &quot;array&quot;,<br>                &quot;items&quot;: {<br>                  &quot;type&quot;: &quot;object&quot;,<br>                  &quot;required&quot;: [&quot;op&quot;, &quot;path&quot;, &quot;value&quot;],<br>                  &quot;properties&quot;: {<br>                    &quot;op&quot;: {<br>                      &quot;type&quot;: &quot;string&quot;,<br>                      &quot;enum&quot;: [&quot;add&quot;]<br>                    },<br>                    &quot;path&quot;: {<br>                      &quot;type&quot;: &quot;string&quot;<br>                    },<br>                    &quot;value&quot;: {<br>                      &quot;type&quot;: [&quot;string&quot;, &quot;integer&quot;]<br>                    }<br>                  }<br>                }<br>              }<br>            }<br>          }<br>        },<br>        &quot;responses&quot;: {<br>          &quot;200&quot;: {<br>            &quot;description&quot;: &quot;Work item created&quot;<br>          },<br>          &quot;default&quot;: {<br>            &quot;description&quot;: &quot;Unexpected error&quot;<br>          }<br>        }<br>      }<br>    }<br>  }<br>}</pre><h3>Step 7 — Define the Agent Instructions</h3><p><strong>Configure agent behavior</strong>:</p><pre>## Role &amp; Greeting<br>At the start of every new conversation, greet the user and introduce yourself as the AI agent that helps create Azure DevOps work items.</pre><pre>## Data Collection<br>Ask the user for four fields:<br>- Title – &quot;What&#39;s the short, descriptive title for the work item?&quot;<br>- Description – &quot;Please give me the detailed description.&quot;<br>- Priority (1–4) – &quot;What priority should we give this? 1 = Critical, 2 = High, 3 = Medium, 4 = Low.&quot;<br>- Tags – &quot;What tags should we add? Use semicolons, e.g., backend;api;urgent.&quot;</pre><pre>## Validation Rules<br>- Title and Description must be non-empty strings<br>- Priority must be integer 1–4 (convert words like &quot;High&quot; to numbers)<br>- Tags must be non-empty string (semicolons optional but recommended)</pre><pre>## Create Work Item<br>After collecting four valid values, call createBugWorkItem with:<br>- System.Title ← Title<br>- System.Description ← Description  <br>- Microsoft.VSTS.Common.Priority ← Priority<br>- System.Tags ← Tags</pre><pre>## Field Path Format<br>When calling createBugWorkItem, ensure all field paths use correct Azure DevOps format:<br>- Title: &quot;/fields/System.Title&quot;<br>- Description: &quot;/fields/System.Description&quot;<br>- Priority: &quot;/fields/Microsoft.VSTS.Common.Priority&quot;  <br>- Tags: &quot;/fields/System.Tags&quot;</pre><pre>The JSON Patch payload must look exactly like this:<br>[<br>  {&quot;op&quot;: &quot;add&quot;, &quot;path&quot;: &quot;/fields/System.Title&quot;, &quot;value&quot;: &quot;&lt;title&gt;&quot;},<br>  {&quot;op&quot;: &quot;add&quot;, &quot;path&quot;: &quot;/fields/System.Description&quot;, &quot;value&quot;: &quot;&lt;description&gt;&quot;},<br>  {&quot;op&quot;: &quot;add&quot;, &quot;path&quot;: &quot;/fields/Microsoft.VSTS.Common.Priority&quot;, &quot;value&quot;: &lt;priority_number&gt;},<br>  {&quot;op&quot;: &quot;add&quot;, &quot;path&quot;: &quot;/fields/System.Tags&quot;, &quot;value&quot;: &quot;&lt;tags&gt;&quot;}<br>]</pre><pre>When the API call succeeds, reply: &quot;✅ Work item #&lt;id&gt; created successfully — view it at &lt;url&gt;.&quot;</pre><pre>Be concise, friendly, and professional. Help users correct invalid input.</pre><p><strong>Save</strong> the agent.</p><h3>Step 8 — Test and Validate</h3><p><strong>Test your complete solution</strong>: <strong>Agent </strong>→ <strong>Try in playground</strong></p><p><strong>Test Scenarios:</strong></p><ol><li>Start conversation — agent should greet and ask for details</li><li>Provide work item information with all four fields</li><li>Verify work item creation in Azure DevOps via API response</li></ol><h3>Step 9 — Troubleshoot (Optional)</h3><p><strong>Manual API Test:</strong></p><pre>curl -X POST \<br>  -H &quot;Authorization: Basic $AUTH&quot; \<br>  -H &quot;Content-Type: application/json-patch+json&quot; \<br>  --data-binary @payload.json \<br>  &quot;https://dev.azure.com/ORG/PROJECT/_apis/wit/workitems/\$Bug?api-version=7.1&quot;</pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9bf76dadd2c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI for the Cloud Practitioner — Part I — Create an AI Foundry agent with a Logic App Action]]></title>
            <link>https://davidokeyode.medium.com/ai-for-the-cloud-practitioner-part-i-create-an-ai-foundry-agent-with-a-logic-app-action-cbbd4a47c37d?source=rss-77eea12d7ea0------2</link>
            <guid isPermaLink="false">https://medium.com/p/cbbd4a47c37d</guid>
            <dc:creator><![CDATA[David Okeyode]]></dc:creator>
            <pubDate>Wed, 16 Jul 2025 22:10:37 GMT</pubDate>
            <atom:updated>2026-03-04T18:49:46.082Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI for the AWS Cloud Practitioner— Create a Bedrock Agent with a Lambda Action Group</h3><p>This post will walk you through the process of creating an AI agent using Amazon Bedrock that uses an AWS Lambda action group to perform actions in DynamoDB. By the end of this post, you’ll have an AI agent that can create support tickets in a DynamoDB table — without writing complex orchestration code.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RUkN_5611bzOHh3QeDO8LA.png" /></figure><h3>Pre-Requisites</h3><p>Before creating your AI agent with a Lambda action group, ensure you have:</p><ul><li>An AWS account with Amazon Bedrock access</li><li>Model access enabled for <strong>Anthropic Claude 3.5 Sonnet</strong> (or your preferred model) in the Bedrock console</li><li>Basic familiarity with the AWS Management Console</li></ul><h3>What We’re Building</h3><p>We’re creating a two-part solution:</p><ol><li><strong>A Lambda Function</strong> that handles business logic — receiving ticket details and writing them to a DynamoDB table</li><li><strong>A Bedrock Agent</strong> that hosts the AI model, collects information from the user, and invokes the Lambda function through an action group</li></ol><h3>Part 1: Creating the Lambda Function and DynamoDB Table</h3><p>A Lambda function that works with Bedrock Agents needs three key components:</p><ul><li>An <strong>event handler</strong> that parses the input from the Bedrock Agent</li><li>The <strong>business logic</strong> (creating the ticket in DynamoDB)</li><li>A <strong>response format</strong> that the agent can interpret</li></ul><h4>Step 1: Create Your DynamoDB Table</h4><ol><li>Navigate to <strong>AWS Console → DynamoDB → Create table</strong></li><li>Configure the table:</li></ol><ul><li><strong>Table name</strong>: SupportTickets</li><li><strong>Partition key</strong>: TicketId (String)</li></ul><p>3. Leave all other settings as default</p><p>4. Click <strong>Create table</strong></p><h4>Step 2: Create Your Lambda Function</h4><ol><li>Navigate to <strong>AWS Console → Lambda → Create function</strong></li><li>Choose these settings:</li></ol><ul><li><strong>Function name</strong>: create-support-ticket</li><li><strong>Runtime</strong>: Python 3.12</li><li><strong>Architecture</strong>: x86_64</li></ul><p>3. Click <strong>Create function</strong></p><h4>Step 3: Add DynamoDB Permissions to the Lambda Role</h4><ol><li>In your Lambda function, go to <strong>Configuration → Permissions</strong></li><li>Click the <strong>Execution role</strong> link to open it in IAM</li><li>Attach the managed policy: AmazonDynamoDBFullAccess (for production, use a scoped-down policy)</li></ol><h4>Step 4: Add the Lambda Function Code</h4><p>Replace the default code with the following, then click <strong>Deploy</strong> to save your function:</p><pre>import json<br>import boto3<br>import uuid<br>from datetime import datetime<br><br>dynamodb = boto3.resource(&#39;dynamodb&#39;)<br>table = dynamodb.Table(&#39;SupportTickets&#39;)<br>def lambda_handler(event, context):<br>    # Extract the function and parameters from the Bedrock Agent event<br>    action_group = event.get(&#39;actionGroup&#39;, &#39;&#39;)<br>    function_name = event.get(&#39;function&#39;, &#39;&#39;)<br>    parameters = event.get(&#39;parameters&#39;, [])<br>    # Parse parameters into a dictionary<br>    param_dict = {}<br>    for param in parameters:<br>        param_dict[param[&#39;name&#39;]] = param[&#39;value&#39;]<br>    # Extract ticket fields<br>    title = param_dict.get(&#39;Title&#39;, &#39;No title provided&#39;)<br>    description = param_dict.get(&#39;Description&#39;, &#39;No description provided&#39;)<br>    priority = param_dict.get(&#39;Priority&#39;, &#39;3&#39;)<br>    assigned_to = param_dict.get(&#39;AssignedTo&#39;, &#39;Unassigned&#39;)<br>    # Generate a unique ticket ID<br>    ticket_id = f&quot;TKT-{str(uuid.uuid4())[:8].upper()}&quot;<br>    # Write the ticket to DynamoDB<br>    ticket_item = {<br>        &#39;TicketId&#39;: ticket_id,<br>        &#39;Title&#39;: title,<br>        &#39;Description&#39;: description,<br>        &#39;Priority&#39;: int(priority),<br>        &#39;AssignedTo&#39;: assigned_to,<br>        &#39;Status&#39;: &#39;Open&#39;,<br>        &#39;CreatedAt&#39;: datetime.utcnow().isoformat()<br>    }<br>    table.put_item(Item=ticket_item)<br>    # Return the response in the format Bedrock Agents expects<br>    response_body = {<br>        &quot;TEXT&quot;: {<br>            &quot;body&quot;: json.dumps({<br>                &quot;message&quot;: f&quot;Support ticket {ticket_id} created successfully!&quot;,<br>                &quot;ticketId&quot;: ticket_id,<br>                &quot;title&quot;: title,<br>                &quot;priority&quot;: int(priority),<br>                &quot;assignedTo&quot;: assigned_to<br>            })<br>        }<br>    }<br>    action_response = {<br>        &quot;actionGroup&quot;: action_group,<br>        &quot;function&quot;: function_name,<br>        &quot;functionResponse&quot;: {<br>            &quot;responseBody&quot;: response_body<br>        }<br>    }<br>    return {<br>        &quot;messageVersion&quot;: &quot;1.0&quot;,<br>        &quot;response&quot;: action_response<br>    }</pre><h4>Step 5: Test Your Lambda Function</h4><ol><li>Click <strong>Test</strong> and create a new test event with this payload:</li></ol><pre>{<br>  &quot;actionGroup&quot;: &quot;TicketActionGroup&quot;,<br>  &quot;function&quot;: &quot;CreateSupportTicket&quot;,<br>  &quot;parameters&quot;: [<br>    { &quot;name&quot;: &quot;Title&quot;, &quot;value&quot;: &quot;Login page broken&quot; },<br>    { &quot;name&quot;: &quot;Description&quot;, &quot;value&quot;: &quot;Users cannot log in after the latest deployment&quot; },<br>    { &quot;name&quot;: &quot;Priority&quot;, &quot;value&quot;: &quot;1&quot; },<br>    { &quot;name&quot;: &quot;AssignedTo&quot;, &quot;value&quot;: &quot;david@example.com&quot; }<br>  ],<br>  &quot;messageVersion&quot;: &quot;1.0&quot;<br>}</pre><p>2. Click <strong>Test</strong> — you should see a success response with the generated ticket ID</p><p>3. Verify the item was created in <strong>DynamoDB → SupportTickets → Explore table items</strong></p><h3>Part 2: Creating the Bedrock Agent</h3><p>Now we’ll create an AI agent that can intelligently collect the required information and call our Lambda function.</p><h4>Step 1: Enable Model Access</h4><ol><li>Go to <strong>Amazon Bedrock → Model access</strong> (left navigation under Bedrock configurations)</li><li>Click <strong>Modify model access</strong></li><li>Select <strong>Anthropic → Claude 3.5 Sonnet</strong> and submit</li><li>Wait for the status to show <strong>Access granted</strong></li></ol><h4>Step 2: Create the Agent</h4><ol><li>Navigate to <strong>Amazon Bedrock → Agents</strong> (left navigation under Builder tools)</li><li>Click <strong>Create Agent</strong></li><li>Configure:</li></ol><ul><li><strong>Agent name</strong>: Support-Ticket-Creator</li><li><strong>Agent description</strong>: A friendly AI that greets users, asks for and validates support ticket details including Title, Description, Priority (1–4), and an assignee email, then creates the ticket in DynamoDB.</li><li><strong>Foundation model</strong>: Anthropic Claude 3.5 Sonnet</li><li><strong>Agent instruction</strong>: (see Step 3 below)</li></ul><p>4. Click <strong>Create</strong></p><h4>Step 3: Set Agent Instructions</h4><p>Copy this instruction set into the <strong>Instructions for the Agent</strong> field:</p><pre>## Role &amp; Greeting<br>On every new conversation, immediately greet the user and introduce yourself<br>as the AI agent that helps create support tickets.<br><br>## Data Collection<br>Ask for these four required fields:<br>- Title (string) – &quot;What&#39;s the short, descriptive title for the ticket?&quot;<br>- Description (string) – &quot;Please give me the detailed description.&quot;<br>- Priority (integer 1-4) – &quot;What priority should we assign?&quot;<br>  1 = Critical, 2 = High, 3 = Medium, 4 = Low<br>- Assigned To (string) – &quot;Who should own this? Provide their email address.&quot;<br>## Validation Rules<br>- Title/Description: must be non-empty strings<br>- Priority: must be integer 1-4 (convert words like &quot;High&quot; to numbers)<br>- Assigned To: must match an email address format<br>## Create Ticket<br>Once all fields are valid, invoke the CreateSupportTicket function with the<br>collected parameters.<br>Be concise, friendly, and professional. Guide users through fixes for<br>invalid entries.</pre><h4>Step 4: Add the Action Group</h4><ol><li>In your agent, scroll down to <strong>Action groups</strong> and click <strong>Add</strong></li><li>Configure:</li></ol><ul><li><strong>Action group name</strong>: TicketActionGroup</li><li><strong>Description</strong>: Creates a support ticket in DynamoDB with the provided details</li><li><strong>Action group type</strong>: Define with function details</li><li><strong>Action group invocation</strong>: Select an existing Lambda function → create-support-ticket</li></ul><p>3. Under <strong>Action group function</strong>, click <strong>Add function</strong> and configure:</p><ul><li><strong>Name</strong>: CreateSupportTicket</li><li><strong>Description</strong>: Creates a new support ticket with the given title, description, priority, and assignee</li></ul><p>4. Add the following <strong>Parameters</strong>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/649/1*CVBnBF8fkQq_el6Q8uykug.png" /></figure><p>5. Click <strong>Create</strong> to save the action group</p><h4>Step 5: Add the Resource-Based Policy to Lambda</h4><p>Amazon Bedrock needs permission to invoke your Lambda function. In the Lambda console:</p><ol><li>Go to <strong>Lambda → create-support-ticket → Configuration → Permissions</strong></li><li>Scroll down to <strong>Resource-based policy statements</strong> and click <strong>Add permissions</strong></li><li>Configure:</li></ol><ul><li><strong>Policy statement</strong>: AllowBedrockInvoke</li><li><strong>Principal</strong>: bedrock.amazonaws.com</li><li><strong>Source ARN</strong>: Your Bedrock Agent ARN</li><li><strong>Action</strong>: lambda:InvokeFunction</li></ul><p>4. Click <strong>Save</strong></p><h4>Step 6: Prepare and Test Your Agent</h4><ol><li>Back in the Bedrock Agent console, click <strong>Prepare</strong> at the top of the page</li><li>Wait for the status to show <strong>Prepared</strong></li><li>In the <strong>Test Agent</strong> panel on the right, start a conversation</li><li>The agent should greet you and ask for ticket details</li><li>Provide the requested information</li><li>Watch as the agent creates the ticket in DynamoDB!</li></ol><h3>Understanding the flow</h3><ol><li><strong>User</strong> sends a message to the Bedrock Agent</li><li>The <strong>Agent</strong> (powered by Claude) interprets the message and determines what information is still needed</li><li>Once all required fields are collected and validated, the Agent invokes the <strong>Action Group</strong></li><li>The Action Group triggers the <strong>Lambda Function</strong> with the collected parameters</li><li>The Lambda Function writes the ticket to <strong>DynamoDB</strong> and returns a success response</li><li>The Agent presents the confirmation to the user</li></ol><p>The beauty of this architecture is the separation of concerns. The Lambda function can be reused by other agents or applications, while the AI agent can be easily modified to change how it interacts with users — all without touching the underlying business logic.</p><h3>References</h3><ul><li><a href="https://docs.aws.amazon.com/bedrock/latest/userguide/agents.html">Amazon Bedrock Agents Documentation</a></li><li><a href="https://docs.aws.amazon.com/bedrock/latest/userguide/agents-action-create.html">Action Groups for Bedrock Agents</a></li><li><a href="https://docs.aws.amazon.com/bedrock/latest/userguide/agents-lambda.html">Configure Lambda Functions for Action Groups</a></li><li><a href="https://docs.aws.amazon.com/bedrock/latest/userguide/agents-action-function.html">Define Function Details for Action Groups</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cbbd4a47c37d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>