What it does

chrome-devtools-mcp is an official Google project that exposes a live Chrome instance to AI coding agents via the Model Context Protocol (MCP). Your agent can click around, take screenshots, read console logs with source-mapped stack traces, record performance traces, and analyze network requests — all through the actual Chrome DevTools frontend, not a toy wrapper. A standalone CLI is available if you want to skip the MCP plumbing.

The interesting bit

The performance tooling is the unusual part: it records DevTools traces and can cross-reference them against Google’s real-user CrUX dataset, giving field data alongside lab metrics. That’s not something you get from a basic headless-browser script. The project also ships “skills” (structured prompts) for some clients, so the agent knows how to use the tools, not just that they exist.

Key highlights

• Built on Puppeteer for automation, but layers on the full DevTools frontend for analysis
• Supports a --slim headless mode for basic tasks without the full DevTools overhead
• Officially supports Google Chrome and Chrome for Testing only; other Chromium browsers are at-your-own-risk
• Opt-out flags for both usage statistics (--no-usage-statistics) and CrUX lookups (--no-performance-crux)
• Installation configs provided for a small army of clients: Claude, Cursor, Copilot/VS Code, Codex, Gemini, Cline, and others

Caveats

• Usage statistics are enabled by default and sent to Google; you must explicitly opt out
• The MCP client gets broad access to anything in the browser instance, so the README warns against feeding it sensitive data
• Update checks phone home to npm by default; disable via environment variable if you’re air-gapped or just allergic to chatter

Verdict

Worth a look if you’re building agentic workflows that need more than static analysis — especially performance debugging or visual regression checks. Skip it if your agent never leaves the terminal, or if you can’t stomach the opt-out-by-default telemetry.