Skip to content
MCP server

Agent Browser

MCP server that lets AI clients drive a real Chromium browser via Vercel's agent-browser

Pricing
Open source
Platforms
CLI, macOS, Linux
Version
0.1.0
Listing updated
today

Similar MCP servers

About Agent Browser

Agent Browser is an open-source MCP server for browser automation. It is available on CLI, macOS, Linux, Windows and self-hosted.

The README describes this as an MCP wrapper around Vercel Labs' agent-browser CLI, so any MCP client can open pages, fill forms, click elements, capture accessibility snapshots and take screenshots. It runs an actual Chromium instance rather than fetching documents, which the README contrasts with the page-fetching tools built into some assistants. A typical flow is to navigate, take an interactive snapshot, then click or fill elements using references from that snapshot.

Session state is kept between runs by default. Cookies and localStorage are saved on close and restored under a namespaced session key, so a login completed in a headed window can carry over after the MCP process restarts. Configuration is done through an optional TOML file that controls headed mode, download location, timeouts, namespacing and an encryption key for restore files.

The project is aimed at developers who want an LLM agent to work through real websites that need a browser, such as logged-in or form-driven flows. The README labels it as beta and notes that restore across a real Chromium restart is still checked manually.

What it does

Real browser control
Runs Chromium through agent-browser so pages with scripts and forms can be operated, not just read.
Accessibility snapshots
Returns the page's accessibility tree with element references that later clicks and fills can target.
Persistent sessions
Saves cookies and localStorage on close so logins can survive a restart when persistence is enabled.
Regex text extraction
Searches page text with a regular expression and returns matches with two lines of context.
Readiness waits
Waits for page load, a URL change, specific text or a selector before the next step.
Diagnostics tool
A browser_doctor call reports on CLI presence, configuration, display and daemon state without launching Chrome.

Good for

  • Logging into a website once in a headed window and reusing that session in later agent runs
  • Filling and submitting web forms as part of an LLM-driven workflow
  • Taking screenshots of pages for an agent to inspect visually
  • Extracting specific text from a rendered page using a regex pattern

Pricing

Open source

MIT-licensed open source, self-hosted.

Agent Browser FAQ

Does it search the web?

No. The README says it does not provide search; you use a search tool to find a URL and then navigate to it with browser_navigate.

What does it need installed?

The README lists installing agent-browser with cargo and running agent-browser install to get the browser, then building the MCP binary from source.

Are saved cookies stored securely?

The README says restore files are credentials and stored in plaintext unless an encryption key file is configured.

Can I watch the browser while it works?

Yes, with the headed setting or AGENT_BROWSER_HEADED=1, though it only takes effect when no headless daemon is already running.

Is Agent Browser free?

Yes. Agent Browser is open source, so you can use and self-host it for free.

What are the best alternatives to Agent Browser?

The closest alternatives to Agent Browser are MCP Browser, playwright_mcp, Chrome DevTools MCP, Agentic Browser and Browsentic.

Where this comes from

  • Description: written from the project's README and its MCP Registry entry.

How AIPoweredBot checks servers