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

# How it works

> Where your data lives, how a request travels from your assistant to Fanaura, and why a new chat sometimes matters.

The Model Context Protocol (MCP) is an open standard that lets an AI assistant call tools on another service. Fanaura runs one MCP server. Your assistant connects to it, signs in as you, and calls Fanaura tools to build journeys and read contacts.

## The pieces

<CardGroup cols={3}>
  <Card title="Your assistant" icon="message-square">
    Claude, ChatGPT, Cursor, or another MCP client. It reads your request, picks a tool, and writes the answer.
  </Card>

  <Card title="The Fanaura MCP server" icon="server">
    `https://app.fanaura.com/mcp`. It checks your sign-in token and runs the tool against your account.
  </Card>

  <Card title="Your Fanaura account" icon="database">
    Journeys, contacts, visits, and answers. This is the only place your data is stored.
  </Card>
</CardGroup>

## A request, step by step

```mermaid theme={null}
sequenceDiagram
  participant You
  participant Assistant
  participant MCP as app.fanaura.com/mcp
  participant Fanaura as Your Fanaura account
  You->>Assistant: "When someone comments GUIDE, DM them my guide"
  Assistant->>MCP: fanaura.create_journey (link, keyword) + sign-in token
  MCP->>MCP: Check the token and find your account
  MCP->>Fanaura: Build the journey, reserve the link, publish
  Fanaura-->>MCP: Journey, public link, recap
  MCP-->>Assistant: live: true, public_url, recap
  Assistant-->>You: "Your GUIDE journey is already live at ..."
```

<Steps>
  <Step title="The assistant chooses a tool">
    Each tool description states what the tool does, what it needs, what it returns, and what it changes. The assistant matches your request to that description. A request with a link and a keyword fits `fanaura.create_journey`. A question about one person fits `fanaura.get_contact`.
  </Step>

  <Step title="Fanaura checks who you are">
    The assistant sends the access token it received when you signed in. Fanaura looks up that token, finds the account you approved, and runs the tool only inside that account.
  </Step>

  <Step title="The tool runs on your live data">
    Nothing is copied or synced ahead of time. Each request reads and writes your Fanaura account at that moment.
  </Step>

  <Step title="The assistant answers from the result">
    Fanaura returns structured data. Journey tools include a recap of the settings. Contact tools include a summary. The description states what came back. It does not tell the assistant how to phrase the answer.
  </Step>
</Steps>

## Where your data lives

| What | Where it is stored | What the assistant keeps |
| - | - | - |
| Journeys, keywords, replies | Your Fanaura account | Nothing after the chat |
| Contacts, visits, answers | Your Fanaura account | Only what appears in that chat |
| Your sign-in | Fanaura issues the token | The access token and refresh token |
| The list of Fanaura tools | Fanaura's server | A copy from the moment you connected |

<Note>
  Your assistant's own chat history follows that assistant's privacy settings. Fanaura does not read your chats. It only sees the tool calls the assistant makes.
</Note>

## Why a new chat sometimes matters

Assistants save the list of Fanaura tools when you connect. When Fanaura adds a tool or improves a description, an existing chat can keep using the older list.

<Tip>
  If the assistant says a Fanaura tool does not exist, or behaves the old way, start a new chat. If that does not help, disconnect and reconnect Fanaura in your assistant's settings.
</Tip>

## One server for every assistant

Fanaura does not run a separate backend per assistant. Claude, ChatGPT, Cursor, VS Code, Claude Code, and Gemini CLI all call `https://app.fanaura.com/mcp`, sign in through the same Fanaura page, and get the same 20 tools. A fix on Fanaura's side reaches every assistant at once.

## Read and write

<Columns cols={2}>
  <Card title="Read tools (12)" icon="eye">
    List journeys, get one journey, check a link, validate publish, setup status, Instagram connect link, audience overview, search contacts, describe contact fields, get contact, contact timeline, and audience activity.
  </Card>

  <Card title="Write tools (8)" icon="pencil">
    Create journey, start journey, set destination, set keyword, set public replies, set capture, set followers only, and publish journey.
  </Card>
</Columns>

Seven of those write tools are marked destructive: create journey, publish journey, set destination, set keyword, set public replies, set capture, and set followers only. Start journey creates a new draft only and is not destructive. The 12 read tools are read only. No tool deletes a journey or a contact. The full list, with every parameter, is in the [tool reference](/mcp-server/tools/index).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.