Rank #07 · Weekly · Agent-Skill
chrome-devtools-mcp: Google's Official Chrome DevTools Bridge for Coding Agents
Point your coding agent at the official Chrome DevTools MCP server (or the bundled CLI) so it can drive a live Chrome browser, inspect network and console data, and record performance traces.
What it does
chrome-devtools-mcp (package chrome-devtools-mcp; README title: Chrome DevTools for agents) is the official Chrome DevTools MCP server and CLI from the ChromeDevTools organization. GitHub description: "Chrome DevTools for coding agents." Docs: developer.chrome.com/docs/devtools/agents.
README claims (as verified):
- Lets a coding agent (examples given: Antigravity, Claude, Cursor, or Copilot) control and inspect a live Chrome browser. It ships as a Model Context Protocol (MCP) server and as a
chrome-devtoolsCLI. - Performance insights: uses Chrome DevTools to record traces and extract actionable performance insights.
- Advanced browser debugging: analyze network requests, take screenshots, and check browser console messages (with source-mapped stack traces).
- Automation: uses Puppeteer to automate actions in Chrome and automatically wait for action results.
- Requirements: Node.js LTS, current stable Google Chrome or newer, and npm. Official support is Google Chrome and Chrome for Testing only; other Chromium-based browsers may work but are not guaranteed.
- Getting started (MCP client config):
npxwithchrome-devtools-mcp@latest. The browser starts when the client uses a tool that needs it — connecting to the MCP server alone does not launch Chrome. - CLI install:
npm i chrome-devtools-mcp@latest -g, then commands such aschrome-devtools navigate_pageandchrome-devtools take_screenshot. See the repo's CLI guide for the full command list. - Optional skills in the package teach an agent how to use the MCP tools, the CLI, and specialized debugging flows (accessibility, cookies, LCP, memory leaks, troubleshooting). They can be installed as a plugin or copied into an agent's skills directory. Skills should be updated whenever you update the server or CLI. This repository is the official DevTools server itself, not a skills-only pack.
- Configuration includes flags such as
--headless,--isolated, and--slim(a reduced tool set for basic browser tasks). Advanced usage covers concurrent sessions, persistent user-data directories, custom browser paths, attaching to a running Chrome, and Android debugging — see the repo guides rather than inventing steps here.
License: Apache-2.0 (LICENSE file and GitHub SPDX agree).
Who it's for
- People running coding agents who want official Chrome DevTools access — performance traces, network/console inspection, screenshots, and Puppeteer-driven automation — over MCP or a CLI.
- Teams standardizing on Chrome / Chrome for Testing rather than a generic multi-engine browser tool.
- Tool builders who want a reference browser subagent on top of Chrome DevTools (the README points to Gemini CLI's browser-agent documentation).
Less ideal if you cannot run current stable Chrome (or Chrome for Testing), need guaranteed support for non-Chrome browsers, or only want a classic Playwright test suite with no agent/MCP loop.
Stars / language / license
58,366 stars · TypeScript · Apache-2.0
Data verified: 2026-11-20 09:04 Asia/Shanghai (official Daily Top10 gate; language/license confirmed via GitHub API get_repository + README + LICENSE)
Trending note: +3,229/wk (weekly Trending window; selection captured 09:04 Asia/Shanghai)
Category tags
Agent-Skill · Chrome-DevTools · MCP · browser-automation · debugging · performance · TypeScript · Google
How it differs from similar projects
Compared with Playwright / Puppeteer / Selenium as test SDKs. Those are general automation libraries. chrome-devtools-mcp packages DevTools inspection and automation for a coding agent, via MCP and a CLI, including traces, console/network debugging, and Puppeteer-backed actions.
Compared with Firecrawl (#01). Firecrawl turns websites into LLM-ready crawl/scrape data. chrome-devtools-mcp is for live browser control and DevTools debugging inside an agent loop. Complementary more often than interchangeable.
Compared with Vercel Labs agent-browser (not on today's Daily board). Both let an agent drive a browser. agent-browser is a Rust CLI built around compact accessibility-tree snapshots and ref handles (@e1). chrome-devtools-mcp is the official Chrome DevTools server, with a stronger documented emphasis on DevTools debugging and performance analysis. They overlap on "open a page and act"; TidyHow has not compared reliability, coverage, or speed, and this is not a claim of feature parity.
Compared with connecting an agent to a raw CDP endpoint. This project wraps DevTools capabilities as MCP tools and a documented CLI, plus client-configuration guides for common agents.
Caveats. The MCP server exposes the browser instance to MCP clients — the README warns you to avoid sharing sensitive data you do not want those clients to see. Usage statistics are enabled by default (opt out with --no-usage-statistics, or CHROME_DEVTOOLS_MCP_NO_USAGE_STATISTICS / CI); this is independent of Chrome's own metrics. Performance tools may send trace URLs to the Google CrUX API unless you pass --no-performance-crux. The server also checks npm for updates unless CHROME_DEVTOOLS_MCP_NO_UPDATE_CHECKS is set. No independent accuracy or speed benchmarks are claimed here.