BrowserTools MCP gives your AI coding agent eyes on the browser you are actually using. It streams console output, network activity, screenshots and Lighthouse audits from your real, logged-in Chrome session to any MCP-compatible client such as Cursor, Claude Code, Windsurf, Cline, Zed or Gemini CLI, so the agent can debug the app you are building with real evidence instead of guesses. The open-source project from AgentDesk AI has about 7,300 stars on GitHub and saw a major 2.0 rewrite in 2026.
Tools that drive a fresh automated browser, such as Playwright-based servers, are great for writing tests but poor at debugging an app mid-development, because they start from a blank session with none of your logins, cookies or current page state. BrowserTools MCP takes the opposite approach: it connects to the Chrome session you already have open, so the agent sees the same console errors, failed network requests and rendered pages you do. Version 2.0 consolidated the architecture into a single process, removed the unauthenticated local server of earlier releases, scrubs credentials before data leaves the browser, and ships with a real test suite. The result is a tighter feedback loop: ask your agent why a page is broken, and it can inspect logs, requests and screenshots itself.
BrowserTools MCP is free and open source under the MIT license. There is no paid tier for the MCP server itself; you run it locally alongside your editor and browser.
BrowserTools MCP is for developers who pair-program with AI coding agents and want those agents to debug front-end issues using the actual state of the app, as well as teams standardizing on MCP tooling across several different AI IDEs.