기본 콘텐츠로 건너뛰기

Chrome DevTools MCP Tutorial: Let Your Agent Debug Chrome

Chrome DevTools MCP (package chrome-devtools-mcp) is an open-source MCP server from the Chrome DevTools team that lets a coding agent open a real Chrome window, read console errors, inspect network requests and record performance traces. This tutorial shows the exact setup for Claude Code and Cursor from the official docs, three first prompts that turn the agent into a front-end debugger, and the settings worth changing before you point it at anything real. It costs nothing to run, and you only need Node.js, npm and a current Chrome.

Why give your coding agent a browser

Most coding agents write front-end code blind. They edit a React component, run the build, and assume the page works because the compiler is happy. The bugs that hurt users, such as a failed fetch, a CORS block, a hydration warning or a 4-second Largest Contentful Paint, only show up when the page actually runs in a browser. The Chrome team's launch post described the problem the same way: agents are "programming with a blindfold on."

Chrome DevTools MCP takes off the blindfold. It speaks the Model Context Protocol, so any MCP-capable client (Claude Code, Cursor, Copilot, Gemini CLI, Codex, Cline, Windsurf and more are listed in the docs) can call DevTools features as tools. Under the hood it uses Puppeteer to drive Chrome and waits for actions to finish, and it pulls in the same performance insights you see in the DevTools Performance panel. The current npm release is 1.13.0, and the repository is licensed under Apache 2.0.

The project has recently been folded into a wider "Chrome DevTools for agents" package that also ships a terminal CLI and agent skills, but the MCP server, the package name and the config below are unchanged.

What you need before you start

  • Node.js LTS and npm. The server runs through npx, so there is nothing to clone.
  • Google Chrome, current stable or newer. The README says it officially supports Chrome and Chrome for Testing only; other Chromium browsers "may work" without guarantees.
  • An MCP client. This guide covers Claude Code and Cursor, but the same JSON works in most clients.

A quick sanity check that works on any OS: run npx -y chrome-devtools-mcp@latest --help in a terminal. On our Linux box with Node 22 it downloaded the package and printed the full list of flags in a couple of seconds. If that command fails, fix Node first; your editor will not do better than your terminal.

Step 1: Add the server to Claude Code

Claude Code has two documented routes. The plain MCP install is one command:

claude mcp add chrome-devtools --scope user npx chrome-devtools-mcp@latest

--scope user makes the server available in every project rather than just the current folder. If you want the bundled skills too (instructions that teach the agent how to chain tools for jobs such as LCP debugging, accessibility audits or memory leaks), install it as a plugin instead, from inside Claude Code:

/plugin marketplace add ChromeDevTools/chrome-devtools-mcp
/plugin install chrome-devtools-mcp@chrome-devtools-plugins

Restart Claude Code afterwards and check that the skills loaded with /skills. The docs warn that if you installed the MCP server earlier, you should remove the old entry first so you don't end up with two copies. If the plugin install fails with Failed to clone repository (common behind corporate firewalls that block git over HTTPS), fall back to the claude mcp add command above, which fetches from npm and skips git entirely, though without the skills.

Step 2: Add the server to Cursor

In Cursor, open Cursor Settings → MCP → New MCP Server and paste the standard config from the README (the docs also offer a one-click install link):

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

The -y flag auto-accepts the npx install prompt, which matters because an MCP client cannot answer it for you. Using @latest means each session picks up new releases; pin a version such as chrome-devtools-mcp@1.13.0 if you prefer predictable behavior over new tools.

The same block works in any client that reads an mcpServers key. On Windows 10, if the client reports MCP error -32000: Connection closed, the troubleshooting guide suggests launching through cmd: set "command": "cmd" and "args": ["/c", "npx", "-y", "chrome-devtools-mcp@latest"].

Step 3: Check that it works

The official first prompt is:

Check the performance of https://developers.chrome.com

Your client should launch Chrome and record a performance trace. One detail that confuses people: connecting to the server does not start a browser. Chrome only opens when the agent calls a tool that needs one, so an idle server with no Chrome window is normal.

The tools your agent gets

The tool reference lists the default set by category: input automation (10 tools such as click, fill_form and upload_file), navigation (6), emulation (2), performance (3), network (2), debugging (10) and memory (14), plus opt-in groups for network mocks, extensions, PWAs, WebAuthn and WebMCP. For front-end debugging, these are the ones that matter most:

  • list_console_messages and get_console_message: every console message since the last navigation, with optional source-mapped stack traces (includeStackTraces).
  • list_network_requests and get_network_request: recent requests, filterable by resource type, with request and response headers and bodies for a single request.
  • performance_start_trace, performance_stop_trace and performance_analyze_insight: record a trace (optionally reloading the page, with or without cache) and drill into insights like LCPBreakdown for Core Web Vitals.
  • take_snapshot and get_css_styles: a text snapshot of the page from the accessibility tree, then the matched CSS rules for an element, ordered by cascade precedence with overridden declarations marked.
  • lighthouse_audit: Lighthouse scores and failing audits for accessibility, SEO and best practices (performance is left to the trace tools).
  • evaluate_script and take_screenshot: run JavaScript in the page and capture what it looks like.

You can see the full category list and our own symptom-to-tool table in the official screenshots at the end of this post.

Your first 3 debugging prompts

Chrome's launch post includes sample prompts such as "A few images on localhost:8080 are not loading. What's happening?" and "Localhost:8080 is loading slowly. Make it load faster." Those work, but agents do better when you name the evidence you want. These are our own versions for a dev server on localhost:3000; adjust the port.

Prompt 1: Fix console errors

Open http://localhost:3000/checkout in Chrome. List all console errors and
warnings with stack traces. For each error, find the source file in this repo,
explain the cause in one sentence, and fix it. Then reload the page and confirm
the console is clean.

This pushes the agent through navigate_page, list_console_messages and back again after the edit, so it verifies its own fix rather than declaring victory.

Prompt 2: Explain failing network requests

Load http://localhost:3000/dashboard and list every network request that
failed or returned 4xx/5xx. For the worst one, show the request URL, method,
status and the response headers, and tell me whether it is a CORS problem,
an auth/header problem or a server bug. Propose a fix in our code.

Asking for headers matters: get_network_request returns them, including Set-Cookie and CORS headers, which is usually where the real answer is.

Prompt 3: Find what slows the page down

Record a performance trace of http://localhost:3000 with a reload and no cache.
Report LCP, CLS and INP if available, pick the single insight with the biggest
impact, analyze it in detail, and make one code change that addresses it.
Record a second trace to compare.

The before-and-after trace keeps the agent honest. Without it, many agents will "optimize" by guessing.

Chrome DevTools MCP vs other browser MCP servers

Chrome DevTools MCP is not the only way to give an agent a browser. The two you will see most often are Microsoft's Playwright MCP and Browserbase's hosted MCP server. They overlap, but they are built for different jobs.

  Chrome DevTools MCP Playwright MCP Browserbase MCP
Price Free, open source (Apache 2.0) Free, open source (Apache 2.0) Hosted service: Free plan $0/mo (1 browser hour), Developer $20/mo (100 browser hours, then $0.12/hr), Startup $99/mo
Where the browser runs Local Chrome (stable, beta, dev or canary), or connect to a running Chrome Local browser driven by Playwright Cloud browsers on Browserbase infrastructure
How the agent sees the page Accessibility-tree snapshot, screenshots, DevTools data Structured accessibility snapshots; no vision model needed Stagehand natural-language tools: navigate, act, observe, extract
Console and network Built in, console with source-mapped stack traces Built in (browser_console_messages, browser_network_requests) Not among the six documented tools
Performance traces Built in, with DevTools insights and optional CrUX field data Tracing is opt-in via --caps=devtools Not a documented tool
Best for Debugging and profiling your own front end Cross-browser automation and test-style flows Scraping and automation at scale, many parallel sessions

The short version: if the question is "why is my page broken or slow," Chrome DevTools MCP gives the agent the deepest view. If the question is "click through this flow reliably," Playwright MCP is lighter. Its own README now points coding agents toward the Playwright CLI with skills to save context. If you need many browsers in the cloud, Browserbase is a paid service built for that.

Editor's note

Who it's for: front-end and full-stack developers who already use Claude Code, Cursor or a similar agent and keep pasting console errors into chat by hand. It also suits anyone chasing Core Web Vitals, because the trace insights are the same ones DevTools shows.

Who it's not for: people who want unattended scraping at scale (a hosted browser service fits better) and teams that cannot use Chrome specifically.

Gotchas we found in the docs:

  • The agent sees everything in that browser. The README warns that MCP clients can inspect and modify any data in the browser instance. Use the default dedicated profile or --isolated (a temporary profile deleted on close) rather than connecting it to your everyday Chrome.
  • Usage statistics are on by default. Add --no-usage-statistics to the args to opt out. Performance tools may also send trace URLs to Google's CrUX API for field data; --no-performance-crux turns that off, which you may want for internal URLs.
  • Target closed error: Chrome could not start. Close any running Chrome instances and make sure the latest stable Chrome is installed.
  • ERR_MODULE_NOT_FOUND: usually an unsupported Node version or a corrupted npx cache. Clear it with npm cache clean --force and reinstall.
  • Running as root in Docker or CI: Chrome won't start as root. Run the server as a non-root user.
  • Sandboxed clients: if your client sandboxes MCP servers, Chrome may fail to launch; either disable the sandbox for this server or start Chrome yourself and connect with --browser-url=http://127.0.0.1:9222.
  • Only the last 3 navigations are kept for console and network data by default (--max-navigation-saved, 1 to 25), so ask the agent to read logs right after reproducing the bug.

FAQ

Is Chrome DevTools MCP free?
Yes. It is an open-source npm package under the Apache 2.0 license and runs on your own machine. You still pay for whatever model your coding agent uses, and every tool result (snapshots, traces, logs) adds to that model's context.

Can it debug a page that is already open in my Chrome?
Yes. Start Chrome with remote debugging and pass --browser-url=http://127.0.0.1:9222, or use --autoConnect with Chrome 144+ after enabling remote debugging at chrome://inspect/#remote-debugging. Chrome requires a non-default user data directory when you enable the remote debugging port, so a separate profile is the safer default anyway.

Does it work headless, for example in CI?
Yes, add --headless to the args. The --slim flag cuts the tool list to three basic tools (navigation, script execution and screenshots) if you only need light automation and want a smaller context.

Related on lightmind24: Claude Code Errors Fixed: Login, PATH, 429 and Usage Limits · Agent Skills: One SKILL.md for Claude Code, Cursor, Copilot · AWS Emulator Tutorial: Let Coding Agents Test AWS Locally

Set it up once, keep the profile isolated, and your agent stops guessing about what the browser is doing.

Screenshots of the official README and Claude Code setup section, the tool reference and the Chrome for Developers launch post are at the end of this post, along with our own symptom-to-tool table. Screenshots: github.com/ChromeDevTools/chrome-devtools-mcp and developer.chrome.com, captured 2026-11-15; tool-by-task table by lightmind24.

Last verified: 2026-11-15
Sources: GitHub: ChromeDevTools/chrome-devtools-mcp · Client configurations guide · Tool reference · Configuration guide · Troubleshooting guide · Chrome for Developers: Chrome DevTools (MCP) for your AI agent · Get started with Chrome DevTools for agents · GitHub: microsoft/playwright-mcp · Browserbase MCP server docs · Browserbase pricing

lightmind24 is an independent review site, not affiliated with Google, Microsoft or Browserbase.

Uploaded Image Uploaded Image Uploaded Image Uploaded Image

댓글

이 블로그의 인기 게시물

브루(Vrew) 요금제 정리: 무료 200크레딧으로 어디까지 쓸 수 있나

요약: 영상 자막 편집기 브루(Vrew)는 2026년 4월 22일부터 기능별 사용량 대신 '통합 크레딧' 하나로 요금을 매겨요. 무료 플랜은 매달 200크레딧이 들어오는데, 음성 분석(자동 자막)만 쓰면 최대 200분 분량이라 짧은 영상 몇 편은 충분하지만, 영상 초반의 Vrew 마크를 지우려면 유료 플랜이 필요해요. 이 글은 브루 공식 가격 페이지와 고객센터 공지를 2026년 10월 23일에 다시 확인해, 크레딧 계산법과 플랜 고르는 기준을 정리했어요. 목차 브루 요금 구조: 통합 크레딧이란 1크레딧으로 할 수 있는 일 무료 플랜으로 되는 것과 안 되는 것 유료 플랜 4종 비교표 월 구독·연 구독·기간 이용권, 무엇이 이득일까 상황별 추천: 학생·직장인·유튜버 크레딧이 모자랄 때: 사용 기록 지우개 업그레이드·환불 전에 알아둘 규칙 운영자 메모, 자주 묻는 질문 1. 브루 요금 구조: 통합 크레딧이란 예전 브루는 플랜마다 음성 분석 몇 분, AI 목소리 몇 자, 번역 몇 자처럼 기능별 한도를 따로 줬어요. 그런데 보이저엑스(브루 운영사)가 실제 이용 데이터를 보니 한두 가지 기능만 집중해서 쓰는 사람이 훨씬 많았다고 해요. 그래서 2026년 4월 22일 업데이트부터는 플랜마다 정해진 크레딧을 받고, 어떤 기능에 쓸지는 내가 정하는 방식으로 바뀌었어요. 플랜별 매달 지급량은 Free 200 · Light 2,000 · Standard 10,000 · Business 12,000크레딧 이에요. 공식 공지에 따르면 무료 사용자는 매달 1일에, 유료 사용자는 매월 결제일에 충전되고, 남은 크레딧은 다음 달로 넘어가지 않아요. 바뀐 뒤 무엇이 좋아졌나 공지에 나온 예시가 이해하기 쉬워요. Standard 플랜에서 AI 목소리만 쓰면 예전에는 최대 50만 자였지만, 지금은 1만 크레딧을 모두 AI 목소리에 몰아 쓸 수 있어 최대 100만 자가 돼요. 반대로 여러 기능을 골고루 쓰는 사람은 크레딧이 기능끼리 ...

에이닷 전화 vs 익시오: 통화 녹음·AI 요약, 내 통신사·폰에 맞는 앱 고르기

요약: 통화 녹음과 AI 요약을 쓰고 싶다면 먼저 내 통신사와 휴대폰 종류부터 확인해야 해요. LG유플러스 고객은 무료 앱 익시오, SKT 고객은 에이닷 전화가 가장 기능이 많고, 다른 통신사의 아이폰 사용자는 iOS에 들어 있는 기본 통화 녹음·전사 기능이 현실적인 선택이에요. 이 글은 2026-10-30에 SKT·LG유플러스·Apple 공식 페이지와 도움말을 직접 확인해 세 가지를 비교했어요. 왜 '앱 기능'보다 '내 조건'을 먼저 봐야 할까요 통화 요약 앱 비교 글을 보면 기능 목록부터 나열하는 경우가 많아요. 그런데 실제로 써 보려는 사람에게 가장 먼저 걸리는 건 기능이 아니라 이용 조건 이에요. 익시오는 LG유플러스 고객 전용이고, 에이닷 전화는 누구나 설치할 수 있지만 통역콜·baro 같은 기능은 SKT 고객에게만 열려 있어요. 아이폰은 운영체제 구조상 앱이 통화를 마음대로 녹음하기 어려워서, 통신사 앱이 안 되면 iOS 기본 기능에 기대야 해요. 그래서 이 글은 "어떤 앱이 제일 좋은가"보다 "내 통신사·내 폰에서 무엇을 쓸 수 있나" 순서로 정리했어요. 기능 차이는 그다음이에요. 한눈에 비교: 에이닷 전화 vs 익시오 vs 아이폰 기본 통화 녹음 구분 에이닷 전화 (SKT) 익시오 ixi-O (LG유플러스) 아이폰 기본 통화 녹음 (Apple) 누가 쓸 수 있나 통신사 구분 없이 앱 설치 가능. 단, 통역콜·baro·그룹통화 발신 등은 SKT 고객 전용 U+ 모바일 고객. 타 통신사, 만 14세 미만, 선불요금제 고객은 제한 (공식 사이트 FAQ상 알뜰폰 중 KB리브엠은 예외) 지원 국가·언어의 iPhone 사용자. 1명과 통화할 때만 녹음 가능 지원 OS·기기 Android 10 이상, iOS 16 이상 iOS 17 이상 iPhone 12 이상 / Android...

타입캐스트 vs 클로바더빙: 수익 나는 채널에 쓸 때 무료·유료 조건 비교

요약 타입캐스트와 클로바더빙은 둘 다 글만 입력하면 한국어 AI 목소리를 만들어 주는 서비스지만, 무료로 쓸 수 있는 범위가 꽤 다릅니다. 클로바더빙 무료는 매달 15,000자까지 쓸 수 있는 대신 수익이 생기는 채널에는 쓸 수 없고, 타입캐스트 무료는 처음에 주는 약 5분 분량의 다운로드로 끝나는 체험판에 가깝습니다. 광고 수익이 붙은 유튜브나 쇼핑몰 영상에 넣을 목소리라면 처음부터 유료 플랜 조건을 보고 고르는 편이 안전합니다. 왜 이 두 서비스를 같이 비교하나 쇼츠 내레이션, 수업 자료 더빙, 상품 소개 영상처럼 "내 목소리 대신 AI 목소리"가 필요할 때 한국 사용자들이 가장 많이 찾는 곳이 네오사피엔스의 타입캐스트 와 네이버의 클로바더빙 입니다. 기능 소개만 보면 비슷해 보이지만, 실제로 문제가 되는 건 음질보다 "이 목소리를 내 채널에 써도 되나" 입니다. 그래서 이 글은 기능 나열보다 무료 사용 범위, 출처 표기 의무, 유료 플랜 가격과 다운로드 한도를 중심으로 정리했습니다. 모든 수치는 2026년 10월 25일에 두 서비스의 공식 요금 페이지와 고객센터 도움말에서 직접 다시 확인한 내용입니다. 한눈에 보는 비교표 항목 타입캐스트 무료 타입캐스트 베이직 클로바더빙 Free 클로바더빙 Standard 월 요금 0원 9,900원 (연간 결제 시 10% 할인) 0원 19,900원 (네이버 클라우드 플랫폼에서 결제) 무료·포함 사용량 음성 생성·재생 무제한, 다운로드는 최초 3,000크레딧(약 5분) 1회 지급 월 30,000크레딧(약 60분) 월 15,000자, 다운로드 월 20회, 프로젝트 누적 5개 (매월 1일 갱신) 월 30,000자, 다운로드 월 40회, 프로젝트 누적 20개 쓸 수 있...