Add Browser MCP to Cursor

Give Cursor's agent control of your real, already-logged-in Chrome — install takes about 60 seconds.

{
  "mcpServers": {
    "browser-mcp": {
      "command": "npx",
      "args": ["@agent360/browser-mcp"]
    }
  }
}

Paste that into ~/.cursor/mcp.json (global — available in every project) or .cursor/mcp.json inside one project (that project only), reload Cursor, then load the Chrome extension once — done. 34 browser tools, your actual cookies and sessions, works on 2FA and CAPTCHA-gated sites where headless tools (Playwright, Puppeteer) get blocked. MIT-licensed, free, and 100% local — nothing leaves your machine.

If you want the full walkthrough, keep reading. If you just needed the config block, that's it above.

Step-by-step install (Cursor)

Step 1 — Run the installer (gets the Chrome extension onto disk)

npx @agent360/browser-mcp install

This copies the Chrome extension files to ~/.browser-mcp/extension/ — the terminal prints the path, copy it, you'll need it in Step 3. As a side effect it also writes a browser-mcp entry into Claude Code's config; harmless to leave in place even if you don't use Claude Code.

Step 2 — Point Cursor at the MCP server

Cursor doesn't read Claude Code's config, so it needs its own entry. Two ways to add it:

A — Edit the config file directly (most reliable):

Open (or create) one of these and add the mcpServers block from the top of this page:

~/.cursor/mcp.json — global, the server is available in every Cursor project
.cursor/mcp.json inside a specific project's folder — that project only

B — Use Cursor's UI: open Cursor Settings (gear icon, top right) → Tools & MCP → New MCP Server. ⚠ VERIFICÉR — the exact menu wording has moved before across Cursor releases; if you don't see it, editing the JSON file directly (Method A) always works. The UI method just opens the same mcp.json for you to fill in.

Step 3 — Load the extension in Chrome

Chrome doesn't let extensions install themselves from npm, so this one step is manual:

Open Chrome and go to chrome://extensions
Toggle Developer mode ON (top right)
Click Load unpacked (top left, next to "Pack extension")
Navigate to ~/.browser-mcp/extension/ and select it
Mac: in the file picker, press Cmd+Shift+G, paste ~/.browser-mcp/extension/, hit Enter
Windows: paste %USERPROFILE%\.browser-mcp\extension\ into the address bar
Linux: type ~/.browser-mcp/extension/ into the path field

Don't want Developer mode on? Use the Chrome Web Store install instead — see below.

Step 4 — Reload Cursor

Reload the window (Cmd/Ctrl+Shift+P → "Reload Window") or fully quit and reopen Cursor — it reads mcp.json on startup and launches the server process automatically. Open Cursor Settings → Tools & MCP and confirm browser-mcp shows as loaded with an active status dot.

Verify it's working

Ask Cursor's agent to navigate to a URL or take a screenshot of the current tab. If it acts instead of saying it has no browser access, you're connected.

Alternative installs

No npm? Manual zip download

Download the extension zip from the latest GitHub release
Unzip it anywhere (e.g. ~/Downloads/browser-mcp-extension/)
Follow Step 3 above, but select the unzipped folder instead of ~/.browser-mcp/extension/
Add the mcpServers block from the top of this page to ~/.cursor/mcp.json (or .cursor/mcp.json for one project) by hand

No Developer mode — Chrome Web Store

Install from the Chrome Web Store →

No Developer mode toggle needed, and Chrome updates the extension automatically in the background. Then add the same mcpServers block to Cursor's config as in Step 2 — the Chrome Web Store install only replaces Step 3, not Step 2.

Works with any MCP client

Browser MCP is a standard stdio MCP server — it doesn't know or care which client is driving it. The same npx @agent360/browser-mcp command works for Cursor, Claude Code, VS Code's agent mode, Windsurf, or any other MCP-compatible client; only the config file it goes into changes. See the Claude Code install if you also use that.

What your agent can do

The 2FA killer move

This is the reason people install Browser MCP: Cursor's agent hits a login wall, needs a verification code, and — because it's driving your actual logged-in Chrome rather than a fresh headless session — it can switch to your own Gmail tab, read the code, and finish the sign-in itself. No API can do that; there's no "read my 2FA code" endpoint to call. It works because Browser MCP isn't simulating a browser, it's operating yours: your cookies, your sessions, your already-passed 2FA challenges.

The same real-session advantage is why it clears CAPTCHA and anti-bot checks that block Playwright and Puppeteer outright — the traffic genuinely is coming from a human-operated Chrome, because it is one.

34 tools

Category	Tools
Navigation & content	`browser_navigate`, `browser_get_page_content`, `browser_screenshot`, `browser_execute_script`
Interaction	`browser_click`, `browser_fill`, `browser_press_key`, `browser_scroll`, `browser_wait`, `browser_hover`, `browser_select_option`, `browser_set_combobox`, `browser_set_date`, `browser_dismiss_overlays`, `browser_handle_dialog`
Tabs & frames	`browser_list_tabs`, `browser_switch_tab`, `browser_close_tab`, `browser_get_new_tab`, `browser_list_frames`, `browser_select_frame`
Data & network	`browser_get_cookies`, `browser_set_cookies`, `browser_get_local_storage`, `browser_set_local_storage`, `browser_fetch`, `browser_wait_for_network`, `browser_extract_token`, `browser_console_logs`, `browser_upload_file`, `browser_drop_file`
CAPTCHA solving	`browser_solve_captcha` — auto-detects reCAPTCHA v2/v3, hCaptcha, Turnstile, FunCaptcha; ~80% pass rate on the checkbox variant, human fallback for the rest
Human-in-the-loop	`browser_ask_user` — overlay dialog for 2FA, CAPTCHA, or credential input, right inside the page
Meta	`browser_about` — session/extension info

browser_extract_token ships with zero-config shortcuts for 9 common services (Stripe, HubSpot, Slack, Shopify, Mailchimp, Pipedrive, Calendly, Google, LinkedIn) — but it isn't limited to them. For any other provider, the agent falls back to browser_navigate + browser_get_page_content to find and extract the token itself.

Full source: github.com/Agent360dk/browser-mcp.

Why this over Playwright MCP

	Browser MCP	Playwright MCP
Browser	Your real Chrome	Headless (fresh session)
Logins/cookies	Already authenticated	Must log in every time
2FA / CAPTCHA / anti-bot sites	Works — it's your session	Frequently blocked
Human-in-the-loop	`browser_ask_user`	None
Multi-session	10 concurrent sessions, color-coded tab groups	Single session
Install	Config block above	`npx @anthropic-ai/mcp-playwright`

Running more than one Cursor conversation at once

Each conversation gets its own MCP server on its own port (9876–9885), and the extension keeps every session's tabs in a separate color-coded Chrome tab group — one conversation can't see or click another's tabs. Idle sessions auto-exit after 4 hours without commands.

FAQ

How do I add an MCP server to Cursor? Add a browser-mcp entry to ~/.cursor/mcp.json (global) or .cursor/mcp.json in a project (project-only) — the block is at the top of this page. Or use the UI: Cursor Settings → Tools & MCP → New MCP Server. Reload the window or restart Cursor afterward so it picks up the new server.

What is Browser MCP? An MCP (Model Context Protocol) server that gives Cursor — or any MCP client, including Claude Code and VS Code agent mode — control of your actual, already-logged-in Chrome: your cookies, your sessions, your 2FA. 34 tools, MIT-licensed, 100% local.

Is it free? Yes. MIT license, no account, no paid tier.

Does it only work with Cursor, or also Claude Code / VS Code? Any MCP-compatible client. It's the exact same server and the exact same config block — {"mcpServers": {"browser-mcp": {"command": "npx", "args": ["@agent360/browser-mcp"]}}} — only the file you paste it into changes per client.

Should I use the global or project config? Global (~/.cursor/mcp.json) if you want Browser MCP available in every Cursor project, which is what most people want. Project-scoped (.cursor/mcp.json inside one repo) if you only want it active there — useful if you're on a team and don't want it turning up in a shared repo's config for everyone else.

Why do I have to load the extension manually instead of it just installing? Chrome blocks extensions from self-installing from npm or any script — that's a Chrome security boundary, not a Browser MCP limitation. Loading unpacked once, or installing from the Chrome Web Store, are the only two ways in.

Does my browsing data leave my machine? No. The MCP server runs locally over stdio, talks to the extension over a local WebSocket, and the extension talks to Chrome through Chrome's own APIs. Nothing is sent to a remote server.

How do I update it? The MCP server updates itself — every run uses npx @agent360/browser-mcp, so there's nothing to pin or bump. The extension auto-updates only if you installed it from the Chrome Web Store; if you loaded it unpacked, re-run npx @agent360/browser-mcp install and click ↻ reload on chrome://extensions.

Chrome extension says "not connected" — what do I check? Confirm it's loaded under chrome://extensions, click the extension icon → "Reconnect," and give it 2–3 seconds — it scans ports 9876–9885 for the running MCP server.

I already run several other MCP servers in Cursor — will Browser MCP's 34 tools be a problem? Cursor limits how many tools can be active across all your MCP servers combined, so if you're already close to that ceiling, disable tools you don't need from Settings → Tools & MCP (you can toggle individual tools per server, not just whole servers).

Is this the same as browsermcp.io? No — different project, same underlying idea (MCP + your real Chrome), separate codebase. If you found this page searching generically for "browser mcp," make sure you're grabbing the one you meant: this one is @agent360/browser-mcp on npm, github.com/Agent360dk/browser-mcp on GitHub.

Can I run it across multiple Cursor windows/conversations at once? Yes — up to 10 concurrent sessions, each on its own port with its own color-coded Chrome tab group, so sessions can't see or control each other's tabs.