# DataFrame MCP

Server URL: `https://dataframe.fit/api/mcp`

Human guide (HTML): https://dataframe.fit/mcp

## Configuration

```json
{
  "mcpServers": {
    "DataFrame.fit": {
      "url": "https://dataframe.fit/api/mcp"
    }
  }
}
```

## How an athlete connects

1. Add the server URL (or the JSON above) in Claude, Cursor, GPT-4, Gemini, Grok, DeepSeek, or another AI assistant that supports a remote MCP server.
2. The assistant opens DataFrame in the browser (OAuth, PKCE).
3. The athlete signs in with Google if needed.
4. They choose permissions for that assistant, then return to the chat.
5. Manage or revoke each assistant under Settings, Connections, Connected AI agents.

One athlete can connect several assistants. Each connection has its own permissions. Revoking one does not revoke the others.

Do not ask the athlete to paste an access token into chat.

## Permissions

Viewing is offered by default. Changes, and push to Garmin, stay off until the athlete ticks them.

- activities:read / activities:write (includes quality-block and per-lap / split times via get_activity_quality)
- plan:read / plan:write
- integrations:write (push to Garmin and sync - separate from editing the plan)
- wellness:read
- analysis:read
- nutrition:read / nutrition:write
- races:read / races:write
- calendar:read
- gear:read
- settings:read / settings:write

Admin tools are never available.

When discussing interval or threshold pace, call `get_activity_quality` - do not use full-session average pace from `get_activity` alone.

## Discovery

- Protected resource metadata: `https://dataframe.fit/.well-known/oauth-protected-resource`
- Authorization server metadata: `https://dataframe.fit/.well-known/oauth-authorization-server`
- Consent page: `https://dataframe.fit/oauth/authorize`
- Token endpoint: `https://dataframe.fit/api/oauth/token`
- Client registration: `https://dataframe.fit/api/oauth/register`

## Related

- ChatGPT guide: https://dataframe.fit/mcp/chatgpt
- Step-by-step: https://dataframe.fit/support/connect-ai-mcp
- Permission details: https://dataframe.fit/support/mcp-scopes
- Site summary: https://dataframe.fit/llms.txt