FansquareAPI reference
Open the app

AI agents

AI agents (MCP)

Connect Claude, ChatGPT or your own agent to your agency data with the Fansquare MCP server.

MCP (Model Context Protocol) is the open standard AI assistants use to call outside tools. The Fansquare MCP server turns the API into tools an agent can call: ask "how much did each creator make this week?" or "tag the fans who spent more than 100 EUR this month", and the agent reads and updates your agency data itself.

Server address

https://api.fansquare.so/mcp

One address for every assistant. The server speaks the Streamable HTTP transport, without sessions: each request stands on its own.

Connect from claude.ai, Claude Desktop or ChatGPT

  1. 1

    Add a custom connector

    Open the connector settings of your assistant (in claude.ai and Claude Desktop, Settings › Connectors › Add custom connector), and paste the server address.

  2. 2

    Sign in to Fansquare

    The assistant opens a Fansquare page. Sign in with the account of the agency owner, choose the access the agent gets (read only, or read and write) and the creators it covers, then approve.

  3. 3

    Ask

    The Fansquare tools are now available in your conversations. Start with a question about your data, the agent picks the tools it needs.

You can disconnect an assistant at any time from Settings › Developer API › API, or from the connector settings of the assistant.

Connect from Claude Code, Cursor or your own agent

Developer tools can authenticate with an API key instead of signing in. Create a key with the access and creators the agent should have, then add the server with the key in the Authorization header.

claude mcp add --transport http fansquare https://api.fansquare.so/mcp \
  --header "Authorization: Bearer $FANSQUARE_API_KEY"

The Fansquare-Version header is not needed: the server always calls the current version of the API.

Tools

Each endpoint of the API is a tool, named after its operation in snake_case: GET /v1/me is get_me, GET /v1/creators is list_creators, PUT …/assignment is assign_fan. A tool takes the path parameters, query parameters and body fields of its endpoint as one set of arguments, and returns the same JSON as the endpoint.

  • The access decides the tools. The agent only sees the tools whose scope its key or sign-in grants: with read-only access, it cannot change anything. A creator outside the access does not exist for the agent, exactly as for the API.

  • Writes ask first. Tools that change data (assign a fan, tag a fan, edit a note) are marked as such, and assistants ask you to confirm before running them.

  • Publications stay in the REST API. An assistant can read the scheduled publications, not create, change or delete them: a post is public on MYM, so it stays a decision made outside the assistant.

  • Errors come back to the agent. A refused call returns the error of the endpoint (status, title and detail, see Errors), so the agent can correct its request.

  • Same limits. Each tool call counts against the rate limit of its endpoint, and the server itself accepts 120 requests per minute per key.

Webhook endpoints are not available as tools: their signing secret is shown once and must never end up in a conversation. Manage them in the app or with the API.