Chrome DevTools MCP
Chrome DevTools for coding agents
About Chrome DevTools MCP
Chrome DevTools MCP is a Model Context Protocol server maintained by the Chrome DevTools team, and it does one thing: it hands your coding agent a real Chrome browser. The agent can navigate to a page, click through it, record a performance trace, read the network requests and the console messages, take a screenshot and emulate a device. That is the difference between an agent that guesses why a page is slow by reading your source and an agent that opens the page, records a trace and reads the numbers off it.
Most agent browser tooling is either a screenshot service or a scripted automation harness. This is neither. It exposes the DevTools panels themselves as tools the agent can call, so the same performance insights, network waterfall and console output you would look at by hand arrive at the model as structured data. Because it comes from the team that owns DevTools rather than a third party wrapping the protocol, it does not drift out of step with Chrome. It is Apache-2.0 licensed, installs from npm, and drives Chrome on your own machine rather than a hosted browser somewhere else.
You add it to your MCP client configuration as a command that runs chrome-devtools-mcp through npx, and the agent picks up the tools it exposes on the next session. From there you describe what you want checked in plain language. The agent launches or attaches to Chrome, navigates, and calls the tools it needs: a performance trace when you ask why something is slow, a network listing when a request is failing, console messages when something throws. A slim mode cuts the tool surface down for agents that only need basic browser control and struggle with a long tool list, and a CLI exists for use outside an MCP client entirely. Configuration snippets are published for Antigravity, Claude Code, Cursor and VS Code, and it ships pre-bundled with Antigravity 2.0.
- •Live Chrome Control - The agent navigates, clicks, fills forms and emulates devices in a real browser rather than a simulation of one
- •Performance Tracing - Record and read a DevTools performance trace, so an agent diagnosing a slow page works from measurements instead of guesses
- •Network Inspection - Read the actual requests and responses to find the failing call rather than reasoning about it from source
- •Console Access - Surface the errors and warnings the page really produced, which is usually the shortest route to the bug
- •Screenshots - Capture what rendered, for the class of problem that only shows up visually
- •Slim Mode - A reduced tool set for agents that only need basic browser control
- •First Party - Maintained by the Chrome DevTools team, so it tracks Chrome rather than lagging behind it
Developers whose agent keeps confidently fixing the wrong thing on a web page because it has never actually seen the page render. It pays off most on performance work and on bugs that only exist in the browser: layout shifts, a request that fails under one specific condition, a runtime error that nothing in the source predicts. It is not a product in its own right and it does nothing without an MCP client, so skip it if you are not already running an agent that speaks MCP. Read the project warning before pointing it at a logged-in session: it exposes browser content to the MCP client, and the performance tools may send trace URLs to the Google CrUX API, which makes it a poor fit for anything sensitive. Only Google Chrome and Chrome for Testing are supported, so Chromium forks are out.
Pricing
No pricing page found
Free and Apache-2.0 licensed. There is no paid tier and no pricing page: it installs from npm and drives Chrome on your own machine.
From the vendor pricing page, 2026-10-08
