Your GEO data inside ChatGPT, Claude and Cursor.

MCP, the Model Context Protocol, is the standard way an AI client asks an outside system for data or an action. A client that supports it discovers the tools a server offers and calls them mid-conversation. People already draft pages and argue priorities in ChatGPT, Claude or Cursor; with MCP, the assistant reads the real audit before it gives you an opinion.

Fourteen, and they are the whole surface. Twelve read: list_projects, get_latest_audit, list_citations, list_ai_visits, list_prompts, list_topics, list_topic_prompts, list_markets, list_suggestions, visibility_by_model, visibility_trend and compare_visibility. Two write: create_topic and create_prompts, which add topics and prompts to what the platform tracks. The client discovers all fourteen on its own.

It is remote: hosted by Citeme, with no package to install, no process to keep alive and no local port to expose. In Claude you add Citeme as a connector and sign in over OAuth, no key to paste. Other clients use the endpoint URL and a Citeme CLI key. For Claude Desktop's local configuration or Claude Code, the server reads the Citeme CLI credentials: one citeme login covers both.

Any client that implements MCP: in practice ChatGPT, Claude, Cursor and n8n, and the list keeps growing because the protocol is open. In n8n, the fourteen tools become nodes in a workflow, so a scheduled run can read your visibility trend and decide what to do with it, instead of someone checking a chart on Monday morning.

Same data, different consumer. CiteAPI is a REST API: you write code against it and handle auth, pagination and shapes yourself. It is the right tool when you are building a dashboard, a script or a product feature. The MCP server is for when the consumer is a model: it describes its own tools, and you never write a request.

Twelve of the fourteen tools only read. create_topic and create_prompts put new topics and prompts under tracking, measured by the next audit like any other. None of them publishes a page, edits your content, deletes data or changes your project settings. Acting on what the assistant finds happens through Nova, with your approval and its execution log.