Your GEO data inside ChatGPT, Claude and Cursor

Nine, and they are the whole surface. They cover the account, the latest audit, citations, AI visits, tracked prompts, recommendations, visibility by engine, visibility over time, and comparison.

It is remote. The endpoint is hosted by Citeme and you connect a client to it, so there is no package to install, no process to keep alive and no local port to expose. That matters more than it sounds: a local MCP server only works on the machine it runs on, which rules out a hosted n8n instance or a teammate on another laptop. A remote MCP server is reachable from anywhere the client can make a network call. The same endpoint serves your desktop assistant, your editor and your automation platform, and it keeps returning the same numbers to all three.

Any client that implements MCP. In practice that means ChatGPT, Claude, Cursor and n8n, and the list keeps growing because the protocol is open rather than vendor specific. That is the point of standardising on it: Citeme did not have to build four integrations, it built one server. n8n is worth calling out separately. Once the server is connected there, the nine tools become nodes in a workflow, so a scheduled run can read your visibility trend and decide what to do with it instead of a human checking a chart on Monday morning.

Same data, different consumer. CiteAPI is a REST API: you write code against it, you handle auth, pagination and shapes yourself, and you get exactly what you asked for. It is the right tool when you are building something, a dashboard, a script, a product feature. The MCP server is for the case where the consumer is a model rather than your code. It describes its own tools, so the assistant knows what is available and when to call it, and you never write a request. If you are integrating Citeme into software, use the API. If you want to ask questions about your GEO in the assistant you already have open, use MCP.

The nine tools listed above read. They list projects, citations, visits, prompts and recommendations, and they return audits, breakdowns, trends and comparisons. None of them publishes a page, edits your content or modifies your project, which is deliberate: an assistant that can quietly rewrite your site is a much bigger decision than one that can read a report. Acting on what it finds is a different part of the platform. Workflows and integrations are where a fix actually ships, with the approval gates and the execution log that belong there. The MCP server is the read side, and it is the side you want pointed at a model.