GEO MCP — audit whether AI can cite a website

What it is

A public read-only MCP server for teams who care about being quoted by AI answer engines rather than merely ranked by classic search. It focuses on the Chinese ecosystem, where the crawler set, the platforms AI actually cites, and the blocking behaviour all differ from the English web.

Other server: AI Customer Service MCP · All MCP endpoints

Endpoint

https://savantcat.cn/mcp-geo

Transport: Streamable HTTP · Public read-only · No API key · Registered as cn.savantcat/geo-cn

Tools (4)

Tool	What it does
`audit_cn_citability`	One crawl runs a six-layer check → 0-100 score, A-F grade, per-item evidence and a prioritised fix list. Tests 13 China-based crawlers (Bytespider, Baiduspider, Sogou, 360, Yisou, PetalBot …), live AI user-agent blocking at the edge, and Chinese slug pitfalls.
`probe_source_pool`	Give one real user question; it probes the actual source distribution in AI answers and tells you whether you are in the pool. Answers the upstream question: not in the pool means you will not be cited, however good the page is.
`score_visibility`	Turns a batch of real AI answers into five indicators (inclusion coverage, share of answers, accuracy, positive share, negative decline) plus a semantic role-weighted score, with a hallucination guard that re-checks whether the URLs AI cited actually open.
`plan_fixes`	Weight-ranked fix plan: why it matters, how to implement, how to verify — hand it to an engineer or a client, then re-run the same tool to confirm the score moves.

How to use it

{
  "mcpServers": {
    "savantcat-geo": { "type": "http", "url": "https://savantcat.cn/mcp-geo" }
  }
}

Then ask your agent to audit a domain, probe a question, or score a set of answers. Tool output is in Chinese; the schema and tool names are English.

Honest boundaries (carried in the tool output as well)

A score is an observation, not a promise of inclusion or citation by any platform.
Platform behaviour changes; every result carries a timestamp and must be re-measured, not trusted indefinitely.
We do not use llms.txt as a ranking claim: it is checked for existence only and is excluded from scoring.
Unreachable citations are flagged for human review rather than asserted to be hallucinations.
Where a number cannot be obtained, it is left blank with the reason stated — never filled in.

Want this run against your site, with a written report?

Website AI visibility (GEO) — audit, fixes, and monthly measurement

Run the free self-check