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

# MCP server

> Query your FixAEO data from Claude, Cursor, or any MCP client.

Point your AI assistant at your own AI visibility data, then just ask.

## What is MCP?

The Model Context Protocol is an open standard for connecting an assistant to a data source. You add a server once in your client's config. The client then discovers what that server can do and calls it on your behalf, mid-conversation. FixAEO ships one: a read-only view of your account, served over HTTP at `https://api.fixaeo.com/api/mcp` and authenticated with the same API key the public API uses.

## What you get

Fourteen tools. Every one is a read of data you already own, and each runs the same code the dashboard runs, so the numbers match what you see in the app.

<CardGroup cols={2}>
  <Card title="Ask in plain language" icon="message">
    "Which competitors out-cite me on Perplexity this month." No SQL, no export, no CSV round-trip.
  </Card>

  <Card title="Real rows, not summaries" icon="table">
    Your assistant gets the underlying JSON, so it can spot movement you'd skim past in a chart.
  </Card>

  <Card title="Same numbers as the app" icon="equals">
    Each tool calls the existing handler in-process. There's no second pipeline to drift.
  </Card>

  <Card title="Cross-source questions" icon="link">
    Google Search Console data and AI-crawler traffic sit beside your visibility data, in one place.
  </Card>
</CardGroup>

## The honest shape of it

<Note>
  The MCP server needs a paid plan. On a free account the connection is refused rather than returning empty results. See [Plans and limits](/plans-and-limits).
</Note>

Three things are worth knowing before you wire it up.

**It's read-only.** There are no write tools. You can't add a brand, create a prompt, trigger a scan, or change a setting through MCP. Everything runs as a GET against your own workspace. If you want to change something, use the app.

**There are no prebuilt prompts.** Some MCP servers ship a prompt library your client shows as slash commands. This one doesn't register any, and it exposes no resources either. Tools are the whole surface. [Use cases](/mcp/use-cases) is the substitute — a list of questions that map cleanly onto the tools.

**Three tools depend on an integration.** `get_search_performance` and `get_seo_aeo_gap` need Search Console wired up. `get_agent_traffic` needs the tracking tag or a server-log connector. All fourteen still appear in your client's tool list — the three just come back empty until the brand is connected, which reads as "no data" rather than "not connected".

<Note>
  A free account can still create an API key and read the public REST API. The paid gate is specific to MCP.
</Note>

## Start here

<CardGroup cols={2}>
  <Card title="Setup" icon="plug" href="/mcp/setup">
    Config blocks for Claude Desktop, Claude Code, and Cursor, plus where the API key comes from.
  </Card>

  <Card title="Tools reference" icon="wrench" href="/mcp/tools">
    All fourteen tools, their parameters, and what each one returns.
  </Card>

  <Card title="Use cases" icon="lightbulb" href="/mcp/use-cases">
    Sixteen questions worth asking once you're connected.
  </Card>

  <Card title="Customer API" icon="code" href="/api/introduction">
    The same data over plain REST, when you want a script rather than an assistant.
  </Card>
</CardGroup>