ChromeDevTools/chrome-devtools-mcp · 57k stars · Beginner · 12 min · 2026-11-16

Let your coding agent debug a live Chrome page with chrome-devtools-mcp

chrome-devtools-mcp gives your coding agent a live Chrome browser it can drive and inspect, so it can record performance traces, read console and network data and automate pages instead of guessing.

TL;DR

chrome-devtools-mcp is an MCP server (and a CLI) that lets agents such as Claude, Cursor or Copilot control and inspect a live Chrome browser. Add one npx entry to your MCP client config, ask it to check the performance of a page, and the agent opens Chrome and records a trace. It needs Node.js LTS, current stable Chrome and npm.

Why this repo matters

Without a browser, a coding agent debugs front-end problems blind. The README says this server gives it "the full power of Chrome DevTools" for automation, debugging and performance analysis: DevTools-based performance traces with actionable insights, network requests, screenshots, console messages with source-mapped stack traces, and Puppeteer-based automation that waits for action results automatically. It is Apache-2.0 licensed.

Best practices

  • Prefer snapshots over screenshots. The take_snapshot tool description says to prefer a text snapshot of the a11y tree over a screenshot, and to always use the latest snapshot.
  • Treat the browser as shared with the agent. The README disclaimer says the server exposes browser content to the MCP client, so avoid sharing sensitive or personal information you do not want the client to see.
  • Isolate parallel sessions. The advanced-usage guide says to pass --isolated when several MCP client sessions should each launch their own temporary Chrome profile.
  • Use --slim only for basic browsing. The README says --slim exposes a reduced tool set for basic browser tasks (the slim reference lists just navigate, evaluate, screenshot), so leave it off for the debugging and performance steps below.
  • Keep skills in step with the server. The README says skills describe the tools of a specific version, so update them whenever you update the server or CLI.
  • Know the data it sends. The README documents usage statistics (on by default; --no-usage-statistics opts out) and CrUX lookups from performance tools (--no-performance-crux disables).

How-to

Where do I start?

1. Add the server to your client

How do I install it? There is nothing to install by hand; the MCP client runs it with npx. Add this to your client's MCP config:

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": ["-y", "chrome-devtools-mcp@latest"]
    }
  }
}

For Claude Code the client-configurations guide gives a one-liner instead: claude mcp add chrome-devtools --scope user npx chrome-devtools-mcp@latest.

Next: Run your first prompt

Try it

Run the first prompt against a page you own. Then ask the agent to take a snapshot, list console errors, and record a performance trace, and compare what it reports with what you see in Chrome DevTools yourself.

Pitfalls

  • Only Google Chrome and Chrome for Testing are officially supported; other Chromium browsers may work but are not guaranteed.
  • Everything in the browser instance is visible to the agent, so do not sign in to sensitive accounts in that profile.
  • Connecting to a running Chrome with --autoConnect exposes all open windows of the selected profile (default profile if you have several).
  • Outdated skills can tell the agent to use tools or parameters that have changed; update them with the server.

Next steps