Claude in Chrome vs Chrome MCP: which browser tool to use

Claude in Chrome vs Chrome MCP is a fair question once you want an agent in your real browser. Both put the agent inside the Chrome you are signed into. They differ in who builds them, which agents can use them and where permissions live. This post compares Claude in Chrome, Anthropic's own extension, with MCP Browser Extension, the open-source MCP server and Chrome extension this site documents (@mehmoodqureshi/chrome-mcp on npm). It also places Chrome DevTools MCP and Playwright MCP next to both.

MCP Browser Extension is not affiliated with, endorsed by or sponsored by Anthropic or Google. Every statement about Claude in Chrome below comes from the Claude Code docs at code.claude.com/docs/en/chrome and Anthropic's help article on getting started with Claude in Chrome, both checked on 2026-10-09.

The short answer

If you use Claude Code with a direct Anthropic plan and sign in with /login, try Claude in Chrome first. It is first-party, it needs no npx server, and it records GIFs. If you use another MCP host, sign in with an API key, reach Claude only through a cloud provider such as Amazon Bedrock, or work in WSL, Claude in Chrome is not available to you, and an MCP browser extension is the alternative. The same is true if you want a deny-all domain allowlist set on the command line and a per-call audit log.

What Claude in Chrome is

Claude in Chrome is a browser extension from Anthropic. Anthropic's help article describes it as an extension "that allows Claude to read, click, and navigate websites alongside you", started from the Chrome side panel, Claude Cowork or Claude Code.

From the Claude Code docs:

• Requirements. You need a direct Anthropic plan (Pro, Max, Team or Enterprise), the Claude in Chrome extension version 1.0.36 or later, and a sign-in with /login.
• Auth limits. If you authenticate with an API key or a token from claude setup-token, Claude Code keeps Chrome integration off, even when you pass --chrome.
• Providers. It is not available through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. If you reach Claude only through one of those, the docs say you need a separate claude.ai account to use it.
• Platforms. It is not supported in Windows Subsystem for Linux (WSL). The docs say it works with Chrome and Microsoft Edge, and that Claude Code also sets up the connection in other Chromium browsers such as Brave, Arc, Vivaldi and Opera. Anthropic's help article is narrower: it says the extension is not supported on other Chromium-based browsers.
• Starting it. Run claude --chrome, or run /chrome in a session to check status, manage permissions or turn on Enabled by default.
• Sessions. It opens new tabs for its tasks and shares your browser's login state. When it meets a login page or CAPTCHA, it pauses and asks you to handle it.
• Permissions. Site-level permissions are inherited from the Chrome extension, and you manage them in the extension's settings.

What MCP Browser Extension is

MCP Browser Extension is an MCP server that runs over stdio, plus a Manifest V3 extension that connects to it through a localhost WebSocket. It never launches a browser of its own. The extension carries out each call inside your running Chrome, so the agent sees what you see, signed in as you. The logged-in Chrome page explains the model.

Because it is a plain MCP server, it works with any MCP host: Claude Code, Claude Desktop, Cursor, Windsurf and others. It does not care how your agent signs in or which model it runs.

Register it with Claude Code:

claude mcp add chrome-mcp -s user -- \
  npx -y @mehmoodqureshi/chrome-mcp \
  --allow-domain example.com --enable-mutations --persist-token

Or add it to any host that reads a JSON config, such as .cursor/mcp.json:

{
  "mcpServers": {
    "chrome-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@mehmoodqureshi/chrome-mcp",
        "--allow-domain",
        "example.com",
        "--enable-mutations",
        "--persist-token"
      ]
    }
  }
}

On Windows, use cmd as the command and put /c and npx at the start of the arguments. Then load the extension. The bundled folder the server unpacks pairs itself on the next server boot. The Chrome Web Store build is the other option, and it can trail the npm package by a version. The quickstart has every step, and the agent setup page has a prompt that lets your agent do the install.

What it adds on top of the session:

• Deny-all, per-domain allowlist. With no flags the agent can read nothing. Each --allow-domain opens one host, and *.example.com covers its subdomains. Reads are gated as well as clicks, and the extension re-checks the policy on its side.
• Opt-in capabilities. Opening tabs, navigating, clicking and typing need --enable-mutations. Without it the agent can only read. Downloads (--enable-downloads), uploads (--enable-uploads) and eval (--unsafe-enable-eval) are separate opt-ins.
• Batch multi-tab. The batch tool runs many calls in one request, in parallel or in series, and each one goes through the same policy gate.
• Audit log. Every call is written to the task's history.jsonl with the URL, the allow or deny verdict, duration, bytes returned and how many secrets were scrubbed.
• Secret handling. Password field values are never returned, and --redact scrubs tokens, API keys and JWTs from page reads.

The security page covers the policy model, and the tools reference lists every tool.

Side by side

A dash means the source docs do not say. It does not mean the feature is missing.

Where Claude in Chrome is the better pick

• It is first-party. Anthropic builds the extension and the Claude Code integration together. Setup lives in /chrome, and Claude Code offers to install the extension when it needs a browser.
• There is no server to run. There is no npx process, no package to update and no pairing token.
• It records GIFs. You can ask Claude to record a flow as a GIF to document or share it. MCP Browser Extension does not do this.
• It works beyond Chrome. The Claude Code docs list Microsoft Edge, and Claude Code also sets up the connection in Brave, Arc, Vivaldi and Opera, though Anthropic's help article says other Chromium browsers are not supported. MCP Browser Extension supports Chrome only.
• It has more surfaces. The same extension runs in the Chrome side panel (in beta) and in Cowork, with saved shortcuts and scheduled tasks according to the help article.
• Teams can manage it centrally. On Team and Enterprise plans, the help article says admins can restrict sites with allowlists and blocklists.

Where an MCP browser extension fits better

• You use more than one agent. Cursor, Windsurf, Claude Desktop and Claude Code can all use one server.
• You sign in with an API key or a setup token. The Claude Code docs say Chrome integration stays off in that case.
• You reach Claude only through Bedrock, Google Cloud or Foundry. Claude in Chrome is not available there without a separate claude.ai account.
• You work in WSL. Claude in Chrome is not supported there. MCP Browser Extension runs on native Windows, so WSL is not needed.
• You want policy in config, not clicks. The allowlist is a set of flags you can read, commit and review.
• You want a record. history.jsonl answers "what did the agent do in my browser" after the fact.

If you searched for a "claude in chrome free alternative": MCP Browser Extension does not need an Anthropic plan of its own. You still need some MCP host and a model to run it, and those have their own terms.

Claude in Chrome vs Playwright MCP and Chrome DevTools MCP

Neither Playwright MCP nor Chrome DevTools MCP is a drop-in replacement for Claude in Chrome. Both start their own browser by default, and both can reach your logged-in browser if you ask.

Playwright MCP, from Microsoft, drives a browser through Playwright and reads pages as accessibility snapshots. Its README lists Chrome, Firefox, WebKit and Edge through --browser, runs headed by default with a --headless option, and uses a persistent profile by default. Its --extension flag connects to a running Chrome or Edge through the Playwright Extension. The README also says Playwright MCP is not a security boundary. Pick it over Claude in Chrome for end-to-end tests, other browsers or headless runs.

claude mcp add playwright npx @playwright/mcp@latest

Chrome DevTools MCP (chrome-devtools-mcp, published under the ChromeDevTools GitHub organization) focuses on debugging: performance traces, network requests, screenshots and console messages with source-mapped stack traces. By default it starts Chrome with its own profile under ~/.cache/chrome-devtools-mcp/. Its --autoConnect flag connects to a running Chrome 144 or later once you enable remote debugging at chrome://inspect/#remote-debugging. Usage statistics are collected by default, and --no-usage-statistics turns them off. Pick it for performance and debugging work.

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

Other extension-based projects, such as hangwin/mcp-chrome and Browser MCP, also use your logged-in browser. The Chrome DevTools MCP vs Playwright MCP post compares all of them in one table, and the compare page covers the policy and audit differences.

Claude in Chrome vs the built-in browser

Most browser MCP servers launch a separate, signed-out browser. Both tools here work in the browser you already use, so the agent starts signed in. If you want a clean, signed-out browser instead, Playwright MCP with --isolated, which keeps the profile in memory, fits better. For background, see what an MCP browser extension is and can AI agents log in to websites.

Use both?

You can. They are separate extensions with separate connections. A practical split:

• Use Claude in Chrome from Claude Code for quick tasks where you want GIF recordings or the side panel.
• Register MCP Browser Extension in Cursor, Claude Desktop or any other host, or in Claude Code sessions that sign in with an API key.
• Keep the MCP Browser Extension allowlist narrow, so a second agent can only reach the domains you named.

Two cautions. Do not let two agents drive the same tab at once. And watch context: Claude Code's docs note that enabling Chrome by default loads its tools in every session, and each MCP server you add also loads its own tools. MCP Browser Extension's --tools flag trims its catalog to the tools you list. Host setup for the MCP route is in the guides for Claude Code and Cursor.

FAQ

Is MCP Browser Extension the same as Claude in Chrome?

No. Claude in Chrome is Anthropic's own extension. MCP Browser Extension is a separate open-source MCP server and extension, not affiliated with Anthropic, that works with any MCP host.

Can I use Claude in Chrome with an API key?

Not with Claude Code. Its docs say Chrome integration stays off when you authenticate with an API key or a claude setup-token token, because the extension cannot authenticate with those credentials.

What is a Claude in Chrome alternative for Cursor?

Any MCP browser server Cursor can load. MCP Browser Extension, hangwin/mcp-chrome and Browser MCP use your logged-in Chrome, and Playwright MCP's --extension mode and Chrome DevTools MCP's --autoConnect can attach to a running browser.

Why can the agent not open a page after setup?

MCP Browser Extension is deny-all by default. Add an --allow-domain for the site, and add --enable-mutations, which opening tabs and navigating require.

Which one records GIFs?

Claude in Chrome. Its docs describe recording browser interactions as GIFs. MCP Browser Extension takes screenshots but does not record GIFs.