@fastergeo/mcp

FasterGEO as a Model Context Protocol server — GEO audits, Brand Entity Funnel metrics, machine-verified tickets, live AI-engine sampling, and the fabrication gate as 9 tools any MCP-capable agent can call.

claude mcp add fastergeo -- npx -y @fastergeo/mcp     # Claude Code
# or point any MCP client at: npx -y @fastergeo/mcp   (stdio)

Tools

| Tool | What it does | |---|---| | list_engines | 18 supported engines (10 China-market + 8 global) with key-configured status — booleans only, never key values | | sample_engine | Ask one question to one engine; verbatim answer + citations + model + channel | | audit_page | Six-dimension GEO score for one URL, fetched exactly as AI crawlers see it (no JS) | | audit_site | Site-level audit: robots.txt AI-crawler policy, sitemap, llms.txt, per-page scores; unreachable pages are named in failedUrls, never silently dropped | | check_ai_crawlers | Which of the 9 visibility-deciding AI crawlers a site blocks | | compute_metrics | Funnel metrics from raw samples: mention rate, top1/top3, SoV, citations, recognition (knows/unknown/confused with quoted evidence) | | generate_tickets | Findings → prioritized tickets with machine-checkable acceptance criteria | | verify_tickets | Re-check acceptance criteria: todo→done on pass, done→regressed when a fix rots | | check_fabrication | Lint a draft against a fact store: unsourced numbers, superlatives, do-not-claim, E-grade facts |

Engine API keys come from the environment (ZHIPUAI_API_KEY, ARK_API_KEY, OPENAI_API_KEY, …) — same conventions as the fastergeo CLI. No keys → auditing and metrics still work; only live sampling needs them.

Every metric is publicly defined and traceable to code — see METHODOLOGY.md. Unmeasured values are returned as null; do not substitute zeros.

Security note

audit_page / audit_site / check_ai_crawlers fetch caller-supplied URLs. Over the default stdio transport this is equivalent to the local user running curl — fine. Do not expose this server over an HTTP transport to untrusted callers: in a shared or remote environment those tools become an internal-network probe (SSRF). Run it in local, trusted environments only.

License

Apache-2.0 · github.com/arronyounging/fastergeo