Cursor browser MCP: let Cursor control your real Chrome

To let Cursor control your real Chrome, run curl -fsSL https://agentchrome.dev/install | sh, load the Agent Chrome extension once, and restart Cursor. The installer adds an agentchrome server to ~/.cursor/mcp.json, and Cursor's agent can then open pages, click, read the console and record video in your signed-in browser.

Agent Chrome is a remote MCP server, so Cursor only needs a URL and an Authorization header. The browser it controls is the Chrome you already use, with your cookies and sessions.

Install with one command

curl -fsSL https://agentchrome.dev/install | sh

It needs Node.js 18 or newer and Chrome 120 or newer. When it finds Cursor (the cursor-agent or cursor command, or a ~/.cursor folder), the installer:

1. Creates your account and saves the key to ~/.config/agentchrome/config.json.
2. Adds agentchrome to ~/.cursor/mcp.json, after saving mcp.json.bak-agentchrome.
3. Links the agentchrome skill into ~/.cursor/skills.
4. Unpacks the extension to ~/.agentchrome/extension and prints a pairing code.

To install for Cursor only, run agentchrome install cursor.

Load the extension once per Chrome profile

1. Open chrome://extensions and turn on Developer mode.
2. Click Load unpacked and choose ~/.agentchrome/extension.
3. Press Sign in to connect in the popup, or paste the pairing code.

Manual configuration

Add this to ~/.cursor/mcp.json. It is the entry the installer writes:

{
  "mcpServers": {
    "agentchrome": {
      "url": "https://agentchrome.dev/mcp",
      "headers": {
        "Authorization": "Bearer <key from agentchrome key>"
      }
    }
  }
}

Keep any servers that are already in mcpServers. Restart Cursor or reload its MCP settings so it picks up the change.

Use OAuth instead of a key

agentchrome install cursor --oauth

This writes the url with no headers block, so no secret sits in mcp.json. The server supports OAuth 2.1 sign-in for MCP clients that offer it. If your Cursor version does not show a sign-in prompt, use the key form above.

Example prompts for Cursor

• Use agentchrome to open localhost:3000, click the new Pricing button and check that the modal opens.
• Reload the page after my change, read the console with read_console_messages, and fix any errors in the component I just edited.
• Open the production site in my Personal profile and screenshot the checkout page at mobile width.
• Find the failing request on this page with read_network_requests and show me the response body.
• Record a short MP4 that shows the new settings page being used, then give me the link.

What you see while Cursor works

• A tab group per session, named and colored, holding every tab the agent opens.
• A pulsing blue border and a status pill with a Stop button.
• An animated cursor with the agent's name. Your own mouse does not move.
• The popup, which lists sessions with Pause and Resume, plus Stop all.

If you press Cancel on Chrome's "started debugging this browser" bar, every agent stops. The optional settings add a prompt before each new site and a block list. History access is off by default.

Notes for Cursor users

• Work in the editor, check in the browser. Because the agent uses your normal Chrome, it can test the app you are building at localhost using the same logged-in session you use by hand.
• Prompts from other agents transfer. The tool names match Claude in Chrome, so a prompt that works in Claude Code works here.
• Recordings. record captures a clean clip without the border or name tag. See recording.
• One server for every agent. Cursor, Claude Code and Codex can share one paired Chrome, each in its own tab group.

Check that it works

After you load the extension, confirm the whole chain before you rely on it:

agentchrome status

You should see your Chrome profile with a green dot and a check mark next to Cursor. Then start a new Cursor session and ask it to call list_browsers. It should report your profile as "online": true. Next ask it to call tabs_context and navigate to https://example.com. A tab group appears in Chrome with a blue border around the page.

If the agent says No browser is connected, Chrome is closed or the profile is not paired. Open Chrome, check that the popup says Connected, or run agentchrome pair for a new code. If a tool says the session is paused, someone pressed Stop or Cancel on the debugging bar, and you can resume from the popup. Opening file:// pages needs Allow access to file URLs in the extension's details page. More fixes are in troubleshooting.

Related pages

• Test your web app with an AI agent
• Debug console and network errors with an AI agent
• Tool reference
• Install Agent Chrome for every agent
• Agent Chrome vs Playwright MCP

Frequently asked questions

How do I add a browser to Cursor?

Run the Agent Chrome installer, load the extension in Chrome, and restart Cursor. The installer writes an agentchrome entry into ~/.cursor/mcp.json that points at https://agentchrome.dev/mcp.

Does Cursor use my real Chrome profile with this?

Yes. The extension runs inside the Chrome profile you paired, so Cursor's agent sees your logged-in sessions. Each agent session works in its own tab group.

Where is the Cursor MCP config file?

At ~/.cursor/mcp.json, in the mcpServers object. The entry has a url and an Authorization header. The installer saves a backup as mcp.json.bak-agentchrome.

Do I need to open a debugging port for Cursor?

No. The extension connects to the server on its own, and Cursor reaches the server over HTTPS. Chrome only shows a "started debugging this browser" bar while an agent is attached.

Can I use it with Cursor and Claude Code together?

Yes. Both connect to the same server and the same paired Chrome. Each session gets its own tab group, and the popup lists both.

How do I remove it from Cursor?

Run agentchrome uninstall cursor. It deletes the agentchrome entry from ~/.cursor/mcp.json and unlinks the skill.

Updated 2026-11-04 · Agent Chrome is open source (MIT) and operated by Circo, Inc.