LaunchRepoShort answer
- Claude in Chrome is Anthropic’s Chrome extension: Claude reads, clicks and types in your signed-in browser, started from the side panel, from Claude Code or from Cowork. It needs a paid Claude plan.
- Chrome DevTools MCP is a free, open-source MCP server from the Chrome DevTools team: any MCP client, such as Claude Code, Codex, Cursor or Gemini CLI, drives Chrome through the DevTools Protocol.
- Choose Claude in Chrome for work beside you in your own browser, with approval modes and safety checks built in. Choose Chrome DevTools MCP for agent-independent work from the terminal, for debugging and for repeatable runs.
- We tested Chrome DevTools MCP hands-on. Claude in Chrome is compared from Anthropic’s documentation only; the reason is below.
What we tested and what we did not
Tested hands-on
- Chrome DevTools MCP 1.10.1 on our local test form, 5 runs per scenario: fields, file upload, a same-origin and a cross-site iframe, a step only a person may finish, restarts.
- Both ways to run it: it starts its own Chrome, or it attaches to a Chrome a person already uses.
- The full results, with Playwright MCP and agent-browser for comparison, are on the Playwright MCP page.
Documented only
- Claude in Chrome needs a paid Claude plan and the extension signed in inside a Chrome profile. Our test rules exclude personal browser profiles and new paid accounts, so we did not run it.
- Every statement in the Claude in Chrome column comes from Anthropic’s pages as read on the date in the sources.
- We therefore make no claim about how well Claude in Chrome fills forms or handles iframes.
Side by side
Each cell says whether it rests on the documentation or on our own test.
| Topic | Claude in Chrome | Chrome DevTools MCP |
|---|---|---|
| What it is | Docs A Chrome extension that lets Claude read, click and navigate websites alongside you. | Docs An MCP server that gives a coding agent control over Chrome and its DevTools. |
| Who drives it | Docs Claude, from the Chrome side panel, from Claude Code (claude --chrome) or from Claude Cowork. | Docs Whatever MCP client you use, from the terminal or the editor. |
| What you need | Docs A paid Claude plan (Pro, Max, Team or Enterprise) and the extension. Claude Code must be signed in with /login; with an API key the Chrome integration stays off. | Docs Node.js, npm and a current Chrome. Free, Apache-2.0 license; your agent’s model costs still apply. |
| Browsers | Docs Chrome. The Claude Code docs also name Edge and say Brave, Arc, Vivaldi and Opera are detected, while the help center says other Chromium browsers are not supported. | Docs Officially Google Chrome and Chrome for Testing. |
| Your logins | Docs Works in your own browser, so every site you are signed in to is reachable. | Tested Found the login in a Chrome a person had opened, after attaching to it. By default it starts its own profile instead. |
| Approvals and safety | Docs Three permission modes from approving each action to skipping approvals, safety classifiers against prompt injection, blocked site categories and actions it never takes. | Docs No approval step of its own; your MCP client’s tool permissions apply. The README warns that browser content is exposed to the MCP client. |
| CAPTCHA or login page | Docs In Claude Code it pauses and asks you to handle it. | Tested No built-in pause; the agent has to stop. A person finished the human step in the same tab and the agent read the result. |
| File uploads | Docs From Claude Code, up to 10 MB per upload in total. | Tested Uploaded intact from an allowed folder; files elsewhere were refused by default. |
| Pages already open | Docs Works on the tab you are on. | Tested Clicked on a page that was open before it connected, but filled fields only after one reload. |
| Debugging | Docs Reads console output, network requests and the DOM. | Docs Performance traces, Lighthouse, memory snapshots, console with source maps and network requests. |
| Extras | Docs Scheduled tasks, recorded workflows in the classic side panel, sessions saved in your Claude account. | Docs Headless mode, isolated temporary profiles, a CLI without MCP. |
| Data | Docs Screenshots of the tabs Claude works in become part of the conversation. | Docs Usage statistics to Google are on by default and can be switched off. |
Logins, CAPTCHAs and handoff
Both tools reach the sites you are signed in to only if they work in your browser. Claude in Chrome does so by design. Chrome DevTools MCP starts its own profile unless you connect it to your Chrome, either with auto-connect in Chrome 144+ and a permission dialog, or through a debugging port on a separate profile.
For CAPTCHAs, two-factor codes and final approvals, plan a handoff. Anthropic documents that Claude in Chrome pauses on login pages and CAPTCHAs when used from Claude Code. With Chrome DevTools MCP, your agent’s instructions have to say where to stop; in our test, a person ticked the human step in the same tab and the agent then read the submitted result.
Do not try to get around a CAPTCHA with either tool. Solve it yourself in the visible window.
In their own words
Official wording on the points that matter most for a decision, copied from the sources below.
| Topic | Official wording | What it means in practice |
|---|---|---|
| What Claude in Chrome is | Claude in Chrome is a browser extension that allows Claude to read, click, and navigate websites alongside you. | Claude acts inside your browser, next to you. |
| Plans | Claude in Chrome is available for all paid plans (Pro, Max, Team, and Enterprise). | No free plan: you need a paid Claude subscription. |
| Browsers (help center) | Claude in Chrome is not supported on other Chromium-based web browsers or mobile devices. | Plan on Google Chrome on a desktop. |
| Browsers (Claude Code docs) | Chrome integration works with Google Chrome and Microsoft Edge. | From Claude Code, Edge works too; the two Anthropic pages differ here. |
| Logins | shares your browser’s login state, so it can access any site you’re already signed into. | Claude can reach every site you are signed in to, so choose the profile with care. |
| Login pages and CAPTCHAs | When Claude encounters a login page or CAPTCHA, it pauses and asks you to handle it manually. | You stay the one who solves them. |
| Upload limit | a single upload can include up to 10 MB of files in total. | Large files need another route. |
| Manual approval mode | Claude pauses and asks for approval before each action. | The strictest mode asks before every action. |
| Main risk | The biggest risk facing browser-using AI tools is prompt injection attacks where malicious instructions hidden in web content (websites, emails, documents) could trick Claude into taking unintended actions. | Text on a web page can try to give Claude instructions. |
| Recommended setup | Use a separate browser profile without access to sensitive accounts (such as banking, healthcare, government). | Use a browser profile without banking, health or government accounts. |
| Chrome DevTools MCP access | exposes content of the browser instance to the MCP clients allowing them to inspect, debug, and modify any data in the browser or DevTools. | Whatever is open in that Chrome is visible to your agent. |
| Chrome DevTools MCP auto-connect | every time the Chrome DevTools MCP server requests a remote debugging session, Chrome will show a dialog to the user and ask for their permission to allow the remote debugging session. | Chrome asks you each time before an agent attaches to your browser. |
Where each one wins
Claude in Chrome wins when
- you want Claude to work next to you in the browser you already use,
- you prefer built-in approval modes and safety classifiers over configuring them yourself,
- you want scheduled browser tasks or workflows you record once,
- you already pay for Claude and mainly work with Claude.
Chrome DevTools MCP wins when
- you use Codex, Cursor, Gemini CLI or several agents, not only Claude,
- you debug your own web app and need traces, Lighthouse and network details,
- you need headless runs, temporary profiles or tests that repeat exactly,
- you do not want a browser extension with broad permissions.
What LaunchRepo uses
- LaunchRepo’s playbooks are written and tested for Chrome DevTools MCP in your own Chrome, so the same setup works with Claude Code, Codex and other agents.
- Its setup guide connects the agent to your Chrome through auto-connect or a dedicated profile, and you decide once, during onboarding, whether it sends final submissions on its own or asks you first.
- Claude in Chrome is not part of that setup; you can still use it for other work in your browser.
Sources
Official pages and repositories, read on September 30, 2026. Quotes are copied word for word; our test results are our own measurements.
FAQ
- What is the difference between Claude in Chrome and Chrome DevTools MCP?
- Claude in Chrome is an extension that lets Claude act inside your own Chrome, with approval modes and safety classifiers. Chrome DevTools MCP is an open-source MCP server that lets any MCP-capable agent control Chrome and use DevTools features such as traces and network inspection.
- Is Chrome DevTools MCP free?
- The server is free and licensed under Apache-2.0. You pay only for the agent or model you connect to it. Claude in Chrome needs a paid Claude plan.
- Can I use Chrome DevTools MCP with Claude Code?
- Yes. Add it as an MCP server in Claude Code. Claude Code can also use Claude in Chrome through the --chrome flag if you are signed in with a paid Claude plan.
- Which one is safer?
- They protect in different places. Claude in Chrome brings permission modes, safety classifiers and blocked site categories. Chrome DevTools MCP relies on your MCP client’s permissions and warns that browser content is exposed to it. For both, use a profile without sensitive accounts.
- Did you test Claude in Chrome?
- No. It needs a paid plan and a signed-in Chrome profile, which our test rules exclude. Its column is based on Anthropic’s documentation. Chrome DevTools MCP was tested hands-on on our local test form.
- Does Claude in Chrome work in Edge or Brave?
- Anthropic’s pages differ. The help center says it is not supported on other Chromium-based browsers; the Claude Code docs say the Chrome integration works with Chrome and Edge and detects Brave, Arc, Vivaldi and Opera. Check the current docs for your browser.
Launch with your own agent
LaunchRepo gives Claude Code, Codex and other coding agents tested playbooks for launch platforms and directories. They run in your own Chrome, and you decide once, during onboarding, whether the agent sends final submissions itself or asks you first.
See pricing