MCP integration

gemini-cua exposes a curated Model Context Protocol server (streamable HTTP) so coding agents can enqueue and monitor browser QA jobs without writing curl.

Endpoint

http://gemini-cua.apps2.povio.dev/mcp/

Transport: Streamable HTTP (FastMCP). Same optional CUA_API_KEY as the REST API.

Tools

• create_browser_test — start a job (query, initial_url, model, highlight_mouse, metadata)
• get_job — fetch status / summary / links
• list_jobs — recent jobs, optional status filter
• get_job_log — agent log text
• cancel_job — cancel a queued job
• wait_for_job — poll until terminal status or timeout (default 600s)

Browser runs can take minutes. Prefer create_browser_test + wait_for_job, or poll get_job yourself. Video/log URLs are on the job payload.

Cursor setup

Add to MCP settings (project or global). Replace the URL/key for your environment:

{
  "mcpServers": {
    "gemini-cua": {
      "url": "http://gemini-cua.apps2.povio.dev/mcp/",
      "headers": {
        "Authorization": "Bearer YOUR_CUA_API_KEY"
      }
    }
  }
}

If auth is disabled on the server, omit headers.

Claude Desktop / other clients

Any client that supports remote MCP over Streamable HTTP can use the same URL. Send the API key as a Bearer token when CUA_API_KEY is configured.

Suggested agent flow

1. create_browser_test(query=..., initial_url=...)
2. wait_for_job(job_id=..., timeout_seconds=600)
3. read summary; optionally get_job_log / open video link

REST equivalent remains at /docs. A2A: /integrations/a2a.