Giving coding agents the keys to Chrome

What it does This project is an MCP server that hands AI coding agents the reins to a live Chrome browser. It exposes the full DevTools toolkit—network monitoring, console logs with source-mapped stacks, screenshots, and performance trace recording—so an agent can inspect and automate what it sees. A --slim mode is available when you only need basic browser tasks without the full instrumentation overhead.

The interesting bit Because this comes from the Chrome DevTools team itself, performance insights are pulled from the real DevTools frontend rather than cobbled together from screenshots. Under the hood it uses Puppeteer for automation, giving agents reliable action-waiting semantics instead of brittle timers.

Key highlights

• Full DevTools integration: network, console, screenshots, and performance traces
• Puppeteer-based automation with built-in waiting for action results
• Standalone CLI for use outside of an MCP client
• --slim mode for lightweight browser automation
• Official support limited to Google Chrome and Chrome for Testing

Caveats

• The MCP client can inspect, debug, and modify any data in the browser; sensitive information should stay out
• Usage statistics are sent to Google by default; disable with --no-usage-statistics or specific environment variables
• Performance tools may share trace URLs with Google’s CrUX API unless you pass --no-performance-crux
• Other Chromium-based browsers are not officially supported and may behave unexpectedly

Verdict Worth a look if you want your agent to stop guessing about frontend bugs and start actually diagnosing them. Avoid it if you can’t stomach default Google telemetry or if your workflow is tied to non-Chrome browsers.