AI agents & MCP

RankNibbler is built to be called by AI agents — ChatGPT, Claude, Perplexity and any MCP-compatible client. 40 authenticated endpoints, all available on every plan. Two ways to wire it in.

OpenAPI spec (custom GPT actions)

Point a custom GPT Action at the machine-readable spec and configure one API key. Users can then just say "audit this site" and the agent calls RankNibbler directly.

GET /api/openapi.json

MCP server

Run our Model Context Protocol server to expose tools like audit_url to any MCP client — Claude Desktop, IDEs and agent frameworks. The server lives in mcp/ in the repo and proxies to the live API with your key.

Why it works well for agents

• One call, structured JSON — no scraping, parsing or browser automation for the agent to do.
• Live endpoints — instant responses, no async task polling to orchestrate.
• Clear errors — consistent status codes and messages the agent can reason about.
• No third-party spend — no endpoint calls a paid per-request data vendor, so an agent can't run up a bill by looping.

Endpoints agents get the most out of

• /api/v1/page-metrics Beta — one call for SEO, accessibility, security and AI-search signals, so an agent doesn't have to fan out across five endpoints.
• /api/v1/citability Beta — score how quotable a page is for answer engines, or score an agent's own draft before it publishes. The API runs the deterministic scorer only: it never calls a model provider.
• /api/v1/page-diff Beta — hand an agent a structured before/after so it can report exactly what a change did.
• /api/v1/usage Beta — let a long-running agent check its own remaining quota and back off.

Endpoints marked Beta are new: live and callable, but their response shape may still change — pin what you parse.

What the API does not return

Worth telling an agent up front, so it doesn't ask: this is an on-page and technical SEO API. There is no backlink data, no keyword search volume, no SERP positions and no competitor rankings. Only Lighthouse, screenshot and waterfall involve a real browser; every other endpoint analyses the served HTML.