🤖 Can ChatGPT Talk to Salesforce? Exploring Salesforce MCP Servers
After connecting Cursor to Salesforce using MCP, I had a question:
👉 If Cursor can interact with Salesforce through MCP, can ChatGPT do the same?
The answer is yes.
And the experience was surprisingly powerful.
Within a few minutes, I was able to query Salesforce records, update Account data, and retrieve Cases — all without logging into Salesforce.
This was my second hands-on experiment while exploring Salesforce Headless 360.
🚀 Salesforce Is Becoming Truly Headless
Traditionally, interacting with Salesforce meant one of two things:
- Logging into Salesforce UI
- Building integrations using APIs
But MCP introduces a third option.
AI assistants can now interact with Salesforce through tools exposed by Salesforce MCP Servers.
Instead of calling APIs directly, the AI discovers available tools and decides which one to use.
Think of it like this:
🧑 User → 🤖 ChatGPT → MCP → ☁️ Salesforce
🔧 Salesforce Doesn’t Provide Just One MCP Server
One thing I learned during this experiment is that Salesforce provides MCP Servers designed for different use cases.
For developers, the most commonly used option is the Salesforce DX MCP Server.
It exposes capabilities such as:
✅ Querying Salesforce data
✅ Retrieving metadata
✅ Accessing org information
✅ Running tests
✅ User operations
These capabilities become available as tools that AI assistants can discover and invoke.
⚙️ Connecting ChatGPT to Salesforce
The setup was surprisingly straightforward.
Step 1: Create a Salesforce External Client App
Before connecting ChatGPT, Salesforce needs to expose an OAuth-enabled endpoint that ChatGPT can use for authentication.
In Salesforce, create an External Client App and enable OAuth.
During setup:
✅ Enable MCP API and Refresh Token scopes
✅ Generate the Client ID and Client Secret
✅ Configure a callback URL (this can be updated later when ChatGPT generates its final callback URL)
These credentials will be used by ChatGPT to authenticate and access Salesforce through the MCP Server.
Step 2: Enable Developer Mode
Inside ChatGPT Settings, navigate to:
Settings → Apps → Advanced Settings
Enable:
✅ Developer Mode
This unlocks the ability to create custom applications and MCP connections.
Step 3: Create a Salesforce MCP App
After enabling Developer Mode, ChatGPT exposes a new option in settings:
➕ Create App
For the MCP Server URL, I used the Salesforce MCP Server endpoint.
[INSERT SCREENSHOT — CREATE APP]
Step 4: Configure Authentication
This is where the Salesforce External Client App and ChatGPT come together.
For authentication, select:
✅ User Defined OAuth Client
Get Jyoti Agarwal’s stories in your inbox
Join Medium for free to get updates from this writer.
This allows ChatGPT to authenticate as a specific Salesforce user and execute operations in that user’s context.
Next:
- Copy the callback URL generated by ChatGPT.
- Add it to the Callback URL section of the Salesforce External Client App.
- Copy the Client ID and Client Secret from Salesforce.
- Paste them into the ChatGPT app configuration.
Step 4: Connect Salesforce
After entering the Client ID and Client Secret and completing authentication, ChatGPT successfully connected to Salesforce.
💡 If You Encounter a redirect_uri_mismatch Error
After configuring the callback URL in Salesforce and attempting to authorize from ChatGPT, you may still see a redirect_uri_mismatch error initially.
Don’t worry — this is usually because the newly added callback URL hasn’t fully propagated within Salesforce yet.
⏳ Wait a few minutes, then return to the ChatGPT app and try connecting again.
🎉 The Fun Part: Talking to Salesforce Through ChatGPT
Once connected, I asked:
Show me all Accounts from my Salesforce org.
ChatGPT automatically:
✅ Selected the appropriate Salesforce tool
✅ Generated the query
✅ Retrieved records from Salesforce
✅ Formatted the results
💡 Tip: Before running any prompts, make sure Salesforce MCP is selected in ChatGPT’s active tools/apps list; otherwise ChatGPT may respond using its general knowledge instead of querying your Salesforce org.
What impressed me most was that I didn’t tell ChatGPT which tool to use.
It figured that out on its own.
✏️ Updating Salesforce Data
Reading data is one thing.
Updating Salesforce records is where things become really interesting.
I asked ChatGPT to update an Account name.
ChatGPT identified the appropriate tool, requested confirmation, and executed the update.
A quick verification in Salesforce confirmed the change.
💡 What Makes This Different?
The real innovation isn’t the query.
It’s the interaction model.
Traditionally:
User → API → Salesforce
With MCP:
User → AI Assistant → Tool Discovery → Salesforce
The AI assistant determines:
- Which tool is needed
- How to invoke it
- What parameters to pass
This creates a much more natural way to interact with business systems.
🔮 Why I’m Excited About Custom MCP Tools
Everything I demonstrated here used standard Salesforce MCP tools.
But the real potential lies in custom MCP tools.
Imagine asking:
Show me customers at risk.
or
Generate a program status summary.
Those capabilities don’t exist out of the box.
But developers can expose them through custom MCP tools.
That’s where Salesforce Headless 360 starts becoming really powerful.
🎓 Key Takeaways
After connecting ChatGPT to Salesforce using MCP:
✅ Salesforce data became accessible through natural language.
✅ ChatGPT automatically discovered and invoked Salesforce tools.
✅ Standard Salesforce operations could be performed without opening Salesforce.
✅ MCP creates a new interaction model for business applications.
Most importantly, this experiment showed me that Salesforce Headless 360 isn’t just about exposing APIs.
It’s about making Salesforce accessible to AI agents.
And I think we’re only scratching the surface of what’s possible.
