QuarterMaster MCP Server

Overview

The KARAT QuarterMaster API doubles as an official Model Context Protocol (MCP) server via fastapi-mcp. Every REST endpoint mounted under /api/v1 (except browser push notifications) is automatically exposed as a typed, structured MCP tool.

LLM agents operating through MCP execute with the exact same permissions and tenancy boundaries as the authenticated user.

Connection Specification

Connecting

Paste the endpoint above into your client. Nothing else is configured — the sign-in flow is discovered from the endpoint itself, and you log in with the same account you use for the web app.

• Claude — Settings → Connectors → Add custom connector. Leave the OAuth client secret empty.
• ChatGPT — Settings → Connectors → Advanced → Developer mode, then add the URL.

Authentication & Multi-Tenancy

An unauthenticated call to the endpoint answers 401 with a WWW-Authenticate header naming this server's protected-resource metadata. Your client follows that to the authorization server, registers itself, and runs authorization code with PKCE. Auth0 authenticates you; the token it issues carries your stable memberships.

POST /mcp                                  → 401 WWW-Authenticate: Bearer resource_metadata="…"
GET  /.well-known/oauth-protected-resource → the authorization server
GET  /.well-known/oauth-authorization-server → authorize, token, register
POST /mcp  Authorization: Bearer …         → tools

Most tools act on one stable, and you do not have to say which:

• Belong to one stable — it is used automatically.
• Belong to several — call set_active_stable once, and every later tool uses it. get_profile lists yours by name.

Tool Catalog & Naming Conventions

48 of the API's operations are exposed, named {action}_{resource} after their OpenAPI operation_id:

• list_horses / create_horse / update_horse — the stable's horses
• list_shows / create_show_entry — horse shows and their entries
• list_events / create_event — vet, farrier, clinics, training
• list_tack, list_vehicles — equipment and transport
• list_sensors / list_current_readings / list_sensor_stats — environmental monitoring
• get_stable_weather — the forecast where the stable is
• get_profile / list_stables / set_active_stable — who you are and where you are acting

Every tool carries a title and behaviour annotations, so a client knows which calls only read and which change something before it runs them.

Not exposed: the AI harness endpoints, system-prompt overrides, background jobs, web search, browser push, sensor provider credentials, and stable administration — machinery for the web app rather than for an assistant.