MCP server
GEOBee ships a built-in Model Context Protocol server, so an AI agent can crawl a site, list past crawls, and read scored results without you leaving the assistant. It's part of the Pro plan.
Endpoint
The server starts automatically when GEOBee launches and stops when you quit — there's no switch to flip. It listens locally at:
- Endpoint —
http://127.0.0.1:17874/mcp - Status pages —
/and/healthon the same host
If port 17874 is busy, GEOBee tries the next few ports automatically, so check
geobee_server_status (or the /health page) to confirm the live
address. Because it binds to 127.0.0.1, the server is reachable only from your own
machine.
Point your client at GEOBee. In an MCP-capable client, add GEOBee as an HTTP
server using the endpoint above. Keep the desktop app running — the tools drive the same
engine and history you see in the UI.
Tools
Four tools are exposed:
| Tool | What it does | Arguments |
|---|---|---|
geobee_server_status | Return the local GEOBee MCP endpoint status. | None. |
geobee_crawl_url | Crawl an http(s) URL — including localhost — and save the result to GEOBee history. | url (required); maxPages (1–5000, default 50); mode (quick or deep, default quick); concurrency (1–20, default 5); timeoutSeconds (1–120, default 15); includePatterns; excludePatterns. |
geobee_list_crawls | List saved GEOBee crawl history. | domain; limit (1–100, default 20). |
geobee_get_crawl | Return a saved crawl summary and rows. With no id or domain, returns the latest crawl. | id; domain; limit (1–500, default 50). |
Good to know
- One crawl at a time.
geobee_crawl_urlis single-flight — start one crawl and let it finish before kicking off the next. - Shared history. Crawls started over MCP are saved into the same projects you see in the app, so you can open, compare, and export them normally.
- Defaults differ from the UI. Over MCP,
maxPagesdefaults to 50 (and allows up to 5000); the desktop crawl bar defaults to 200.