Floyd MCP Server

MCP server that exposes Floyd scheduling and booking tools to AI agents. Built on the Model Context Protocol Streamable HTTP transport.

For full API documentation, see docs.floyd.run.

Tools

Tool	Description
`floyd_get_available_slots`	Check available appointment times for a service
`floyd_hold_booking`	Place a temporary hold on a time slot
`floyd_confirm_booking`	Confirm a held booking (requires explicit user consent)
`floyd_cancel_booking`	Cancel a held or confirmed booking
`floyd_get_booking`	Retrieve booking details and status

Setup

pnpm install

Environment variables

Variable	Required	Default	Description
`FLOYD_BASE_URL`	No	`https://api.floyd.run/v1`	Floyd API base URL
`PORT`	No	`3000`	HTTP server port

Usage

HTTP mode (Streamable HTTP)

pnpm dev

MCP endpoint: POST / Health check: GET /health

Stdio mode

pnpm dev -- --stdio

Authentication

When the Floyd engine requires authentication, connecting agents provide their API key via:

Authorization: Bearer <token> header (preferred)
?token=<value> query parameter

The API key is forwarded to the Floyd engine on each request. If the engine has auth disabled (e.g. self-hosted), no key is needed.

Booking flow

get_available_slots → hold_booking → confirm_booking
                                   ↘ cancel_booking

Get slots — query available times, returns signed slotId tokens
Hold — reserve a slot using the slotId (or explicit fields). Creates a booking in held state
Confirm — finalize the booking. Requires userConfirmed: true
Cancel — release a held or confirmed booking

Development

pnpm test          # run tests
pnpm typecheck     # type-check
pnpm lint          # eslint
pnpm format        # prettier
pnpm build         # bundle with tsup

Production

pnpm build
pnpm start

The server handles SIGTERM/SIGINT for graceful shutdown. HTTP requests to the Floyd API have a 10-second timeout.

Floyd