One of the fastest-rising repositories on GitHub today is one of Google's own: ChromeDevTools/chrome-devtools-mcp, which added roughly 670 stars in a day and sits at about 55,000 overall.
The project is deliberately unglamorous. It is a Model Context Protocol server — plus a matching command-line tool — that connects a coding agent (Antigravity, Claude, Cursor, Copilot, Gemini CLI) to a live Chrome instance. Instead of asking a model to imagine what the browser is doing, the agent records performance traces, reads network requests, takes screenshots, checks console messages with source-mapped stack traces, and drives pages through Puppeteer.
Google has spent the past few weeks making it easier to adopt. The package now ships "skills" — small instruction bundles that teach an agent how to use the tools for specific jobs, including an accessibility audit, cookie and session debugging, Largest Contentful Paint optimisation and memory-leak hunting. Plugins install the server and every skill in one command for Claude Code, Copilot and the Gemini CLI, and a `--slim` flag exposes a reduced tool set for simple tasks.
The wiring is worth reading before you plug it in. The README is blunt that the server "exposes content of the browser instance to the MCP clients" — anything in a logged-in session becomes readable to whatever agent is attached. Google also collects usage statistics by default (opt out with `--no-usage-statistics`, or by setting `CI`), and performance tools may send trace URLs to the CrUX API for real-user field data unless `--no-performance-crux` is set. Official support covers Google Chrome and Chrome for Testing only; other Chromium browsers may work but are not guaranteed.
Sources
- github.comChromeDevTools/chrome-devtools-mcp — Chrome DevTools for coding agents
- blog.sagarmajumdar.inGoogle Expands Chrome DevTools MCP Server to Equip AI Coding Agents with Direct In-Browser Runtime Debugging and Diagnostics
- mcptools.siChrome DevTools MCP — official registry entry (refreshed 2026-11-09)




