Let Your AI Agent Give Your Website a Checkup with Google Chrome DevTools MCP

Share

Google's official Chrome DevTools MCP lets an AI agent record page performance, check LCP, INP and CLS, run Lighthouse and point to problem elements, take memory snapshots to find leaks, and save screen recordings as mp4. It also ships a command-line tool for scheduled jobs. Includes a prompt that has your AI agent check your site and compare before and after.

Machine translation. Read the original

SourceGitHub · ChromeDevTools/chrome-devtools-mcpReported Nov 11, 2026Not yet tested

Pages load slowly, and they keep using more memory the longer you leave them open. Until now, you had to open DevTools yourself and go through one panel at a time. Google's official Chrome DevTools MCP can now hand that work to your AI agent. It records a performance trace of the page, runs a Lighthouse audit and a memory snapshot, then points out the problems it finds, down to the specific page element responsible. Every feature below was checked against the official repo's tool docs.

What it can check for you

We covered its logpoints and network mocks before. Those are for debugging bugs. This post covers the checkup side only.

What you want to know Tool it uses What it gives back
Why a page loads slowly or doesn't respond to clicks Performance recording (trace) LCP, INP, and CLS, the three core metrics, plus the specific causes slowing them down. You can ask for more detail on any one of them
How the page scores on accessibility, SEO, and best practices Lighthouse audit Scores for each category, the checks that failed, and the page elements they point to. The full report is saved locally
Why memory keeps growing Memory snapshot Which kinds of objects take up the most space, what changed between two snapshots, and what is holding on to them
Keeping a record of the process to share Screen recording Saved directly as an mp4

Note that the Lighthouse audit does not include a performance score. For performance, use the performance recording in the first row. The memory snapshot tools are off by default, so start the server with --memoryDebugging=true to turn them on.

Run it without a chat window

The same package includes a command-line tool, chrome-devtools. Everything above can be called from it, and it runs in the background without a visible window by default. That means you can write a script, put it on a schedule, and have your site checked every day. When something fails, you can ask your AI agent to look into it. The official repo also ships a companion Skill that teaches an AI agent how to write scripts with these commands.

Getting started

Give this prompt to your AI agent:

Connect Google's official Chrome DevTools MCP (GitHub: ChromeDevTools/chrome-devtools-mcp) and run a checkup on my own site: [URL]

1. Add it to my MCP config by following the official README. Before you change anything, tell me which config file you will edit.
2. Run the checks twice, once in mobile mode and once in desktop mode. For each, first record a page load performance trace and list LCP, INP, CLS, and the three biggest causes holding them back. Then run a Lighthouse audit and list the failed checks and the page elements they point to.
3. For each issue, explain how much it matters, which file to change, and how to change it. Show me the plan first, and don't touch any code until I agree.
4. After the changes, test again and put the before and after numbers side by side in one table.
5. Only test my own site. Don't sign in to any account, and don't enter real personal information on the page.

If you want it to run every day, ask your agent to use the chrome-devtools command line to write a script that saves the reports to a fixed folder.

Things to know

  • It can see everything in the browser. The official docs warn that it can read and change any data in the browser it controls. Don't sign in to accounts you don't want an AI to see in that browser.
  • URLs are sent to Google. By default, the performance tools send the URL to Google's CrUX API to pull real-user experience data for comparison. To turn this off, add --no-performance-crux.
  • The command line can read and write every file by default. Use --workspace=<folder> to limit it to your project directory.
  • Chrome only. The official project only guarantees support for Chrome and Chrome for Testing. Other Chromium-based browsers may not work. You need the LTS version of Node.js.

GitHub · Tool reference · CLI docs

Done with this one? Mark it to track your path.Sign in
Save it for laterSign in
Questions about this piece?Ask AI about this

Comments (0)

No comments yet. Start the conversation.

Sign in to join the discussion.Sign in