MCP server · Browser + DevTools
Chrome DevTools MCP
The Chrome DevTools team's MCP server: DOM snapshots, console, network and performance traces from a live Chrome as structured tool results.
Quick verdict
The browser integration for MCP-capable coding agents. Let it launch its own Chrome, keep it off your daily profile, and it belongs in your config.
- MaintainerChrome DevTools team (Google) (official)
- Security scopeLocal execution
- Transportsstdio
- Install channelnpm · chrome-devtools-mcp
- EvidenceHands-on Verified · 1 configuration
- LicenseApache-2.0 · Open source
The official MCP server from the Chrome DevTools team. It exposes a real Chrome to an agent through the DevTools protocol — page navigation and clicks, DOM and accessibility snapshots, console messages, network requests, screenshots, and performance traces with insights — as structured tool results rather than pixels to guess at. By default the server launches its own Chrome with a dedicated profile the first time a tool needs a browser; to attach to a Chrome you already run, enable remote debugging at chrome://inspect/#remote-debugging (Chrome 144+) and pass --autoConnect, or start Chrome with --remote-debugging-port and pass --browser-url. The README documents install steps for Claude Code, Cline, Cursor, Codex, Gemini CLI, Windsurf and many more.
Install Chrome DevTools MCP
npx -y chrome-devtools-mcp@latest
{
"mcpServers": {
"chrome-devtools": {
"command": "npx",
"args": ["-y", "chrome-devtools-mcp@latest"]
}
}
}Node.js LTS and a current stable Chrome. Launches its own Chrome instance with a dedicated profile by default; --autoConnect or --browser-url attach to a browser you already run.
Credentials
No credentials required.
Verified configurations
Each row is one client + runtime + OS combination that was actually run, with what was verified in that session.
| Configuration | Install | Auth | Discovery | Execution | Read | Write | Destructive | Errors |
|---|---|---|---|---|---|---|---|---|
| Claude Code | Verified | N/A | Verified | Verified | Verified | Not tested | Not tested | Partial |
Evidence — Claude Code
- Install: `claude mcp list` → `chrome-devtools: npx chrome-devtools-mcp@latest --isolated=true - ✔ Connected`: the package resolved through npx and Claude Code spawned the stdio process; `--isolated=true` gives the launched Chrome a temporary profile, the posture the profile recommends.
- Auth: The server declares no credentials.
- Discovery: tools/list exposed 29 tools: navigate_page, new_page, list_pages, select_page, close_page, take_snapshot, take_screenshot, click, fill, fill_form, hover, drag, press_key, type_text, upload_file, handle_dialog, wait_for, evaluate_script, list_console_messages, get_console_message, list_network_requests, get_network_request, performance_start_trace, performance_stop_trace, performance_analyze_insight, lighthouse_audit, take_heapsnapshot, emulate, resize_page.
- Execution: `new_page` https://opensourcesai.com/mcp/playwright/ → the server launched its own Chrome and reported page 2 'Playwright MCP | OpenSourcesAI' selected; `take_snapshot` returned the a11y tree with uids (breadcrumb, hero badges, the install block with the config entry, footer); `list_console_messages` returned 3 'Deprecated feature used' issues; `list_pages` listed both pages and `close_page(2)` closed it.
- Read: `take_snapshot` and `list_console_messages` read live page state from the launched Chrome.
- Errors: After `close_page(2)` closed the selected page the server answered with a clean message ('The selected page has been closed. Call list_pages to see open pages.') rather than crashing; no bad-input probe beyond that.
The server launched its own Chrome (default mode) — no --autoConnect / --browser-url attach to a daily browser was tried. Performance tracing and Lighthouse were not exercised.
Best for
Frontend and full-stack agent workflows where the agent should see the real DOM, console errors and network waterfall after a change instead of a screenshot loop.
Why use it
It is the only browser-debugging MCP server from the browser vendor itself, so it tracks the DevTools protocol at the source. Typed DOM, console and network data closes the loop between an agent's code change and what the browser actually did, and performance traces come back with insights rather than raw dumps.
Tradeoffs
It drives a real browser on your machine. Attaching to a Chrome you already use means opening a debugging port that any local application can connect to — keep that to test profiles, never a signed-in daily browser. Chrome-only; no Firefox or Safari.
Upstream state
npm package published · repository active · in the official MCP Registry (active). Lifecycle is derived from a committed upstream snapshot at build time, never authored by hand.
Official MCP Registry name: io.github.ChromeDevTools/chrome-devtools-mcp
Editorial review
Grade A — Hands-on Verified. Re-review is due by 2026-12-16. Config entries, credential names and package status can change upstream; verify against the source repository before granting an agent access.
Sources
Related MCP servers
- ClickHouse MCPQuery ClickHouse databases through the official MCP server, with read-only SQL enforced by default and separate write opt-ins.
- Playwright MCPMicrosoft's MCP server that drives a real browser through Playwright with accessibility snapshots, not screenshots — navigate, click, type, read.
- Cloudflare API MCPSearch and operate the Cloudflare API through a compact Code Mode MCP server with OAuth-scoped account access.