Browser MCP lets your agent work in the Chrome you’re already signed into.
A free, open-source Chrome extension and MCP server. Your agent uses your real browser, with your logins, in its own tab group. It is set up to stop and ask you when it hits a 2FA code, a CAPTCHA or a choice only you can make.
claude mcp add --scope user browser-mcp -- npx @agent360/browser-mcp@latestThe extension does nothing on its own. This command adds the server it talks to.
Free, MIT-licensed, no account. Works with Claude Code, Cursor, Codex, VS Code and Windsurf.
What changes when your agent has your Chrome
It asks you instead of guessing
The agent is instructed to call browser_ask_user when it needs a code, a password or a decision from you. A box appears in that tab, usually with a short chime. You answer, and it carries on.
Each agent session gets its own tab group
Each session opens its own named, colour-coded tab group and works only in its own tabs. There is room for up to 20 sessions at once, one local port each, against the same signed-in Chrome.
Browser MCP uses eight of Chrome’s group colours. After the eighth they repeat; the names don’t. Running several sessions in one Chrome.
It works where you’re already logged in
Your cookies, your sessions, your 2FA. When a sign-in emails a one-time code, it can open Gmail in the same signed-in browser, read the code and finish the sign-in. A fresh headless browser starts without any of that.
How it reads a code from GmailYou say“Log me in at app.example.com. If it emails a code, open Gmail, read the code and continue.”
Install
Browser MCP is two halves: a Chrome extension and a local MCP server. You need both, and Node.js 20 or newer.
Add the MCP server, then restart Claude Code
Paste this in a terminal. Restarting is what starts the server.
claude mcp add --scope user browser-mcp -- npx @agent360/browser-mcp@latestSay this to check it worked
Paste it to Claude Code. You get a screenshot back instead of “I don’t have browser access”, and you’re running. The extension shows no badge until your agent first uses the browser; then a green badge with the number of connected agents appears.
Open example.com and take a screenshot.
Using another client? Same server, that client’s own config: Cursor, VS Code, Codex, Windsurf and more in the guides. No Chrome Web Store? The guides cover the unpacked install too.
Try these first
“Open Gmail and tell me who sent my last 3 emails.”
This one shows the difference. It uses the Gmail you’re already signed into in this Chrome; a fresh headless browser would hit a login wall.
“Go to my analytics dashboard, pull this month’s numbers, and put them in a table.”
A dashboard you’re signed into in this Chrome. No API key, no export, no integration to build first.
“Fill in the signup form at example.com/signup with my details. Stop and ask me before anything sensitive.”
The agent works the page while you stay in the loop, and it can pause to ask you before anything it shouldn’t decide alone.
“Walk through my app’s signup flow as a real user and tell me where it breaks.”
End-to-end QA in the same browser your users have: cookies, extensions, session and all.
It is strongest where you’re signed in and there is no API: internal dashboards, admin panels, portals. CAPTCHAs and bot checks can still stop it: what it can and cannot get past.
How it compares
| Feature | Browser MCP (this) | Playwright MCP | BrowserMCP.io |
|---|---|---|---|
| Browser | Your real Chrome | Own profile by default; your real Chrome via their extension | Your real Chrome |
| Already logged in | Yes | Yes, in extension mode | Yes |
| CAPTCHA assistance | Checkbox click, screenshot, then you | None | Not documented |
| Multi-session | Up to 20 | One tab group per client | Not documented |
| Human-in-the-loop | browser_ask_user | None | Not documented |
| Token-finding guides | 9 providers | None | Not documented |
| Tool count | 40 tools | 72 (counted 2026-09-19) | Not published |
| Install | claude mcp add --scope user browser-mcp -- npx @agent360/browser-mcp@latest | npx @playwright/mcp | npm + extension |
Questions
I installed the extension from the Chrome Web Store and it says “not connected”. What now?
claude mcp add --scope user browser-mcp -- npx @agent360/browser-mcp@latest - then restart it. The popup keeps saying "Not connected" until you ask your agent to use the browser for the first time - that is when the server takes a port, and that is when the green badge appears. The server runs on your machine on ports 9876-9895; the extension scans those ports to find it.What is Browser MCP?
How is this different from Playwright MCP?
browser_ask_user, that puts a question in front of you (a 2FA code, a CAPTCHA, a choice) and waits for your answer. Playwright MCP's tool list has nothing that asks you.Is my data safe? What does the extension see?
127.0.0.1; that link has no password, so it is only as private as your computer (the privacy page says more). The extension and the server have no telemetry and no analytics. What your agent reads is handed to your AI client, which sends it to its model provider like anything else you show it. The cookie tools run only when your agent calls them, but an ordinary page read can contain sensitive text too, such as a token shown on a page, and that text goes to your AI client. Source is open and auditable on GitHub.Why does it need so many permissions?
tabs/scripting for navigation and clicking, cookies for the cookie tools, debugger for real input events on CSP-protected sites like Google Cloud, <all_urls> because the AI must navigate to whatever URL you ask it to. A full per-permission breakdown is on the permissions & data page.Does it work with anything other than Claude Code?
How does CAPTCHA assistance actually work?
browser_solve_captcha first detects the kind (reCAPTCHA, hCaptcha, Turnstile); the agent can then ask it to click a reCAPTCHA v2 checkbox, which is often enough when you are signed into Google. (2) If a challenge is left, the agent can take a screenshot itself and tell the tool which grid cells to click, so the model you already use does any image reading. (3) If that does not clear it, the agent can ask you with browser_ask_user and carry on after you solve it. No CAPTCHA-solving service is involved. It does not pass every CAPTCHA: what it can and cannot get past says which we have measured.Is it really free? What's the catch?
How do I uninstall it?
chrome://extensions, and drop the browser-mcp entry from your MCP config (claude mcp remove --scope user browser-mcp, or delete the block from your Cursor/VS Code/Codex config). Nothing was installed globally - npx only caches the package - and a small local log may sit in ~/.browser-mcp/ (the unpacked install also copies the extension there). Delete that folder to remove those files; the npm cache keeps its own copy of the package. Full details & permissions.Give your agent the Chrome you’re already signed into.
claude mcp add --scope user browser-mcp -- npx @agent360/browser-mcp@latestFree, MIT-licensed, no account. Nothing is sent to Agent360: privacy.