BrowserTools MCP
Console, network, screenshots and Lighthouse audits from your open Chrome session to the agent via a DevTools extension
Install
{
"mcpServers": {
"browser-tools": {
"command": "npx",
"args": ["-y", "@agentdeskai/browser-tools-mcp@latest"]
}
}
}On Windows use command cmd with args /c npx -y @agentdeskai/browser-tools-mcp@latest.
This is third-party code. Review the repository files before installing.
What it does
The server and a Chrome extension pass the agent what happens in a tab with DevTools open: console logs, errors, network requests with bodies, the selected element and storage. The agent can take screenshots, reload the tab and run Lighthouse-based accessibility, performance, SEO and best-practice audits. Unlike automation servers it uses your regular session where you are already logged in. Large payloads are exposed as MCP resources, and auth headers and tokens are scrubbed before they leave the browser.
Who it is for. For frontend developers debugging an app in the browser together with an agent.
Good fit when
- You want the agent to see console errors and failed requests on a live page
- Debugging needs a logged-in session that is hard to reproduce
- You need a quick accessibility or performance audit of a page
Not a fit when
- You need to write and run automated tests in a clean browser
- You use Firefox: support is unverified
Example request
Check the console errors on the open page and see which API requests failedLimitations
Requires Node 22.19 or newer and an extension loaded manually in developer mode. Network capture starts only once DevTools is open. Audits launch a separate Chromium and take up to a minute. Version 1.2.x has a critical vulnerability; upgrade to 2.0.
How to disable. Remove the browser-tools block from the MCP config and remove the extension at chrome://extensions.
MCP
- Transport
- stdio
- Authentication
- not required
| Environment variables | |
|---|---|
| CHROME_PATH | Path to the Chromium-based browser used for audits |
Security check
- Reads data from a logged-in browser session, including request bodies and storage
- Requires an extension with DevTools access
README in short
The README opens with a note about the 2.0 rewrite and the 1.x vulnerability. Setup is an npx MCP config plus manually loading the extension. It has a tool table and covers multiple tabs, resources for large data and pagination parameters. Security measures, launch flags and known limits get their own sections. MIT licensed.
FAQ
How is it different from Playwright MCP?
Playwright MCP drives a separate automated browser, while this server reads your current Chrome session with its logins.
Do I need to start a second server?
Not in version 2.0; the MCP server runs the connector itself.
Related
Official shadcn/ui skill and MCP server: the agent searches registries, adds components via the CLI and follows the library's rules
A frontend design skill with 23 commands, live browser iteration and 61 detector rules for common AI-generated UI tells
Vercel Agent Skills
Agent Skills
Vercel's skills: React and Next.js performance, UI guideline audits, React Native, component composition and deploys
Three skill plugins for Google Stitch: screen generation, code to design and React builds from mockups