Cursor meets a CAPTCHA

There is no honest way to automate past a challenge designed to stop automation. So we do not try to. The agent can try the simple case, and ask you to solve the rest: browser_ask_user brings the tab forward and waits for your answer.

browser_solve_captcha can try the reCAPTCHA checkbox. When a reCAPTCHA escalates to an image grid, the agent can click the cells it chooses after its own screenshot. For a slider or anything else it can call browser_ask_user, which brings the tab forward, shows you a dialog and waits for your answer. browser_ask_user is the same mechanism for anything else only a human can answer.

Setting it up

1 - The Chrome extension. One click from the Chrome Web Store.

2 - The MCP server. Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (one project):

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

Full walkthrough: Install for Cursor.

What we will not tell you

We publish no solve rate. We have not benchmarked it rigorously enough to stand behind a number, and a number without a measurement behind it is worse than no number. What we can say is what the tool does: the agent can try the reCAPTCHA checkbox and image grid, and ask you for the rest.

It will not get you past a challenge that is working. If a site has decided you look like a robot, the honest outcome is that you solve it yourself. The value here is that the run does not die - it pauses, you click, it continues in the same tab with the same session.

Two things to know before you rely on it

The tab has to be in front for the challenge to reach you. Chrome does not deliver mouse and keyboard to a tab that is not the visible one in its window, and the tool brings the tab forward before it asks. That is deliberate: the alternative is a prompt for a challenge you cannot see.

Your session is what makes most of this unnecessary. Many CAPTCHAs fire because a fresh headless browser looks like a bot. It is your real Chrome with your real profile, so on a lot of sites the challenge never appears. That is not a guarantee, and it is not unique to us - Playwright MCP's extension and Chrome DevTools MCP can also drive a signed-in browser.

Frequently asked questions

Does it solve CAPTCHAs for me? It can try the reCAPTCHA checkbox and click the reCAPTCHA grid cells your agent chooses. Anything it cannot clear, your agent hands to you with browser_ask_user. We are not a solving service and do not use one.

Do you send the challenge anywhere? Not to us, and not to a solving service. The server is local, there is no account and no telemetry. If your agent takes a screenshot to choose grid cells, that image goes to your AI client like any other tool result.

Can I skip the attempt and just be asked straight away? Yes - call browser_ask_user yourself, or tell the agent to show you the page instead of trying.

Which other clients does this work with? All of them. Claude Code, Codex, VS Code, Cline, Gemini CLI, Zed, Kiro, Continue.dev.