Chrome DevTools MCP lets a coding agent do what you would otherwise do with DevTools open. It can record a performance trace and explain what slows a page down, list failed network requests, read console errors with source-mapped stack traces, and run a Lighthouse audit. That turns "why is this page slow?" or "why does the submit button do nothing?" into something the agent can investigate and then check again after it changes the code.
It suits front-end developers who want the agent to verify its own work in a real browser.
From the publisher
Official Chrome DevTools MCP server that gives agents control of a live Chrome instance via Puppeteer. Agents can automate input, inspect network requests and console messages, record performance traces with DevTools insights, run Lighthouse audits and analyze heap snapshots.
What you need
- Node.js (an LTS release), npm and Chrome at the current stable version or newer. Google officially supports only Chrome and Chrome for Testing; other Chromium browsers may work.
- No account or API key. It is a local (stdio) server started with
npx. - A client that can start local servers. The project documents Codex setup (
codex mcp add chrome-devtools -- npx chrome-devtools-mcp@latest); on Windows it also needs a longer startup timeout and a couple of environment variables. The Codex CLI, IDE extension and ChatGPT desktop app share one config. - Nothing to launch by hand. The server starts Chrome with its own profile the first time a tool needs a browser.
Good to know
- The agent sees everything in that browser. Google's README warns that the server lets clients inspect and modify any data in the browser. Don't use it to browse sites with personal or sensitive data.
- The profile persists. The default profile in
~/.cache/chrome-devtools-mcp/isn't cleared between runs. Add--isolatedfor a temporary one. - Data goes to Google by default. Usage statistics are on unless you pass
--no-usage-statistics, and performance tools may send trace URLs to the Chrome UX Report API (--no-performance-cruxturns that off). - Script execution can be switched off.
--no-javascript-evaluationdisablesevaluate_script, though the docs stress it isn't a sandbox. - Lighthouse here skips performance. For performance numbers the agent records a trace instead.
- Fewer tools if you want them.
--slimcuts the tool list to three: navigation, script execution and screenshots.
Also available as
Connect
- 1Add the server to Codex
codex mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest
Configs are generated from the publisher's documented endpoint. Check their docs for plan requirements and scopes. New to this? Read the step-by-step tutorial.
Tools 12
navigate_pagenew_pagelist_pagesclickfill_formtake_snapshottake_screenshotevaluate_scriptlist_console_messageslist_network_requestsperformance_start_tracelighthouse_auditThings to ask
- “Check the LCP of web.dev and suggest fixes”
- “Why does the submit button on localhost:8080 not work? Check console errors”
- “List failed network requests on this page”
- “Run a Lighthouse audit on my landing page”
Remove it
- 1In Codex
codex mcp remove chrome-devtoolsOr delete the [mcp_servers.chrome-devtools] block from ~/.codex/config.toml.
Chrome DevTools MCP: common questions
- Can I use Chrome DevTools MCP in ChatGPT?
In Codex, yes, including Codex in the ChatGPT desktop app. ChatGPT on the web only connects to MCP servers over a public HTTPS endpoint or Secure MCP Tunnel, and this server runs as a local command that drives a browser on your machine, so it isn't a good fit there. See Use plugins and MCP servers in Codex for adding a stdio server.
- Can it use my normal Chrome profile and logins?
Not by default; it starts Chrome with a dedicated profile. With
--autoConnect(Chrome 144 or later) it attaches to a Chrome you already have open after you enable remote debugging and allow the connection, and it then has access to every open window in that profile. That helps on sites that block sign-in from automated browsers, but it also exposes your real sessions.- Does Chrome DevTools MCP send data to Google?
By default, yes: Google collects usage statistics such as tool success rates, latency and environment details, separately from Chrome's own metrics. Opt out with
--no-usage-statistics. Performance tools may also send trace URLs to the Chrome UX Report API for real-user data;--no-performance-cruxdisables that.- Chrome DevTools MCP or Playwright MCP?
Pick Chrome DevTools MCP for performance traces, Lighthouse audits and deep debugging in Chrome. Pick Playwright MCP when you need Firefox or WebKit, or mainly want reliable clicking through user flows. They are independent, so you can add both to the same client.
Guide updated Oct 24, 2026. Sources: github.com, github.com, github.com, github.com, github.com, learn.chatgpt.com, developers.openai.com.