> ## Documentation Index
> Fetch the complete documentation index at: https://geoptie.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Use Geoptie from Claude Code, Cursor, Codex and other AI assistants.

The Geoptie MCP server lets an AI assistant work with your AI-search visibility data
directly, using the API key you already have. Ask a question in your own words and the
assistant calls Geoptie for you.

It speaks the [Model Context Protocol](https://modelcontextprotocol.io/), so any client that
supports remote MCP servers connects the same way.

<CardGroup cols={2}>
  <Card title="Endpoint" icon="link">
    `https://mcp.geoptie.com/mcp`
  </Card>

  <Card title="Setup" icon="plug" href="/docs/mcp/setup">
    Connect your client in a couple of minutes
  </Card>
</CardGroup>

## What you can ask for

* Which of our prompts lost visibility this month, and on which engines?
* Which pages do AI engines cite most when they answer about us?
* How does our share of voice compare to our competitors over the last 30 days?
* Why did ChatGPT answer this prompt the way it did, and who else got named?
* What are the open recommendations for this brand, and which look most worthwhile?
* Start tracking these five prompts, and file them under a new topic.
* Write me an article for the gap in this recommendation.

[Recipes](/docs/mcp/recipes) has each of these written out with the tools behind it and how to
read the answer.

## What it covers

Everything the dashboard covers: visibility and share of voice across the engines you track,
the prompts behind those numbers, the full text of the answers engines gave, the pages and
domains they cited, competitors, recommendations, page audits, and content generation.

It reads and it acts. Tracking prompts, managing topics, creating brands, running audits and
writing content are all available, not just reporting.

## How it works

Each tool call is one request to the [Geoptie API](/docs/api-reference) made with your key, so
everything true of the API is true here.

* **The same key.** No separate credential and no separate signup. See
  [Authentication](/docs/authentication).
* **The same access rules.** Your subscription is checked on every request, and a
  brand-scoped key reaches exactly the brands it is scoped to.
* **The same rate limits.** 120 requests per minute per key, plus the per-endpoint limits.
  Anything that runs a model is limited to a few calls per minute and two at a time, so a
  loop stops almost immediately. See [Rate limits](/docs/rate-limits).
* **The same numbers.** You get the values the dashboard shows, not raw rows.

Two things follow from that, and they are worth knowing before you start:

* **Nothing pages by itself.** List tools return one page and a cursor, so your assistant asks
  again for the next one. That is deliberate: it keeps every request counted, which is what
  makes the limits above hold.
* **Anything that runs a model returns a status to poll**, rather than holding the connection
  open for the minutes it takes. Content briefs, drafts and audits all work this way.

Your key is checked when the client connects, so **Connected** means the key is valid and your
subscription is active. You find that out at that moment rather than part-way through a
question.

## What you need

* A Geoptie subscription on an active or trialing plan, on a plan that includes API access
* An API key, created in the dashboard under **API keys**
* A client that supports remote MCP servers, such as Claude Code, Cursor or Codex

## Next steps

<CardGroup cols={3}>
  <Card title="Setup" icon="plug" href="/docs/mcp/setup">
    Connect your client
  </Card>

  <Card title="Tools" icon="wrench" href="/docs/mcp/tools">
    Everything the server exposes
  </Card>

  <Card title="Recipes" icon="list-check" href="/docs/mcp/recipes">
    Questions worth asking
  </Card>
</CardGroup>
