Retriever AI Browser Agent: Free Guide to Scraping and Job Applications
Retriever AI's free browser agent scrapes sites, fills job applications, and learns your workflow through skill files. Here's how it actually works.
What is Retriever AI’s browser agent?
Retriever AI is a browser extension and cloud agent that carries out multi-step tasks on the web from a plain-language instruction. You describe the outcome (a spreadsheet of scraped data, a filled job application, a worked homework problem), point it at the relevant pages, and it navigates, reads, and fills forms on your behalf. It now runs in a free mode that covers AI and cloud browser time for eligible tasks, with a small sponsored card shown alongside results and daily usage limits in place.
TL;DR
- Free mode lets you run Retriever’s browser extension and cloud agent without spending credits, though a sponsored ad appears with results and fair-use limits cap daily usage.
- The agent works through normal-language prompts, such as asking it to build a comparison table from several product pages or find and apply to jobs matching a resume.
- Skill files are markdown documents the agent writes and updates as it learns your preferences, so corrections you make once (like leaving blank cells instead of guessing) carry over to future tasks.
- Beyond personal skills, Retriever maintains shared website skills that capture how to navigate specific sites, improved by successful runs across users and updated by the team as sites change.
- MCP server support lets the agent pull in outside tools (like a database), and a separate MCP feature lets other agents control the Retriever extension itself, though that remote-control option requires a paid plan.
- Scheduled and triggered tasks (running automatically every morning, or firing off an event) stay behind a subscription, while manually starting a task in free mode does not.
- At the time of recording, the starter paid plan was listed at $9.99 a month with a monthly credit allowance for tasks and features beyond the free tier.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
How do you set up Retriever AI’s browser agent?
Getting started is a short process: install the official Chrome extension, sign in, and enable free mode. From there you can hand the agent a task directly from whatever page you have open, which matters if you’re working with sites you’re already logged into (job boards, internal tools, subscription services). Retriever also offers a cloud dashboard as an alternative entry point, useful if you’d rather manage tasks outside the browser itself.
Once it’s running, the agent takes instructions the way you’d brief a person: what to do, which pages to use, and what the final result should look like. Vague prompts get vague results, so specificity does the heavy lifting. A prompt that names the exact columns you want in a table, or the exact criteria for a job search, produces a far more usable output than a general “find me some options” request.
What can you actually use it for?
Three practical use cases stand out.
Web scraping and comparison tables. If you’re comparing products across multiple sites, you can ask the agent to pull specific fields (model, price, connection type, product link) into a table and leave anything it can’t find blank rather than guessing. This same pattern extends to software pricing pages, company directories, or any list of tools you’re researching: swap the pages and the columns, and you have a new task. The catch is that scraped data still needs a human check. Product variants, sale prices versus list prices, and mismatched entries are the kinds of errors worth verifying before you trust the table.
Job applications. You can hand the agent a resume and a target (junior front-end roles, remote-friendly, in your location), and it will search, compare postings against your criteria, explain why a role fits, and start filling out an application. Telling it to pause before submitting keeps you in control of the final send, which matters most for written answers about your experience where the agent’s phrasing may need editing.
Studying and homework help. The agent can read a question on a page, work through a solution, and explain a step you’re stuck on, then generate a similar practice question and check your answer. This turns whatever material you already have open into a study loop without switching tools.
What are skill files and why do they matter?
The feature that separates Retriever from a one-off automation script is memory. The agent keeps personal skill files, markdown documents that capture your preferences and recurring workflows, and it updates them when it learns something durable. If you tell it every comparison table needs a source link and blank cells for missing prices, that preference gets saved and applied automatically the next time you run a similar task. You can open the skills library, read these files, and edit them directly, which keeps the system transparent instead of acting as a black box.
One coffee. One working app.
You bring the idea. Remy manages the project.
Separately, Retriever maintains shared website skills according to the source video: knowledge about how to navigate a specific site, refined by successful runs across users and updated by the Retriever team as sites change their layout. The example given is a filter buried inside a menu that the agent might initially miss. Once a site skill captures the correct path, future runs on that site benefit without needing to relearn it. Personal skills describe how you want to work; site skills describe how a particular website works. Fewer repeated corrections is the point, though the agent still has to apply the saved guidance correctly each time.
How does MCP integration extend the agent?
MCP (Model Context Protocol) is a way for an AI agent to access external tools built by other services, like querying a database or connecting to an internal system. Retriever supports adding MCP servers through the extension’s tool settings: you provide the server’s URL and authenticate if required, and its tools become available during a task. A practical example is comparing information found on a website against records already stored in a connected database.
There’s a second MCP-related capability worth separating out because it affects pricing: you can expose the Retriever browser extension itself through MCP, letting another AI agent control your browser through Retriever. That remote-control setup requires a subscription, unlike connecting outside tools for Retriever to use, which is a distinct and separately available capability.
Is Retriever AI worth paying for?
Manually starting a supported task in free mode costs nothing beyond the sponsored card and daily limits. What stays behind a paywall is automation you don’t have to trigger yourself: background schedules and triggers, like checking a website every morning or kicking off a task when an event happens. The starter paid plan was listed at $9.99 a month with a monthly credit allowance at the time of the source video.
A reasonable way to evaluate it: start with a task you already do by hand regularly (scraping a set of pages, checking job listings, comparing pricing) and run it in free mode first. If the agent handles it reliably and you find yourself wanting it to run unattended on a schedule, that’s the point where paying starts to make sense. If you’re mostly running occasional, manual tasks, free mode with skill files may cover what you need without a subscription.
Frequently Asked Questions
Is Retriever AI actually free to use?
Its free mode covers AI and cloud browser time for eligible manually-started tasks without spending credits, though a sponsored card appears with results and daily fair-use limits apply. Extras like enrichment beyond the included allowance, along with scheduled or triggered automation and remote MCP control of the extension, require a paid plan.
Can Retriever AI fill out job applications for me?
Yes. You can give it a resume and criteria (role type, remote eligibility, location) and it will search for matching postings, explain how each fits, and prepare an application for review. It supports pausing before final submission, which is worth using so you can check written answers before they go out.
What are skill files in Retriever AI?
They’re markdown files the agent writes and updates to capture your preferences and recurring workflows, viewable and editable in a skills library. There’s also a separate category of shared website skills that store navigation knowledge about specific sites, improved through successful runs and maintained by the Retriever team.
Does Retriever AI work with other tools like databases?
Yes, through MCP server connections. You add a server’s URL in the extension’s tool settings and authenticate if needed, after which its tools become available to the agent during tasks, such as comparing web data against records in a connected database.
How much does the paid version of Retriever AI cost?
At the time of the source video, the starter plan was listed at $9.99 a month with a monthly credit allowance for usage beyond what free mode covers, including scheduled and triggered automation.
