Zenoti MCP Server

A Model Context Protocol (MCP) server for the Zenoti spa/wellness/medspa management platform. Exposes appointments, guests, service booking, invoices, catalog, and forms as tools an AI agent can call.

Hosted version available

Do not want to manage credentials, hosting, and updates yourself? DOCK is the managed version of this server: encrypted per-practice auth, audit logs, draft-first write actions, and a workflow library. Founding practices lock lifetime pricing: Front Desk $49/mo (Zenoti), Billing Desk $99/mo (Tebra, BAA included), Full Practice $129/mo (both). https://dockhq.vercel.app

Installation

npm install -g zenoti-mcp-server

Or run from source:

git clone https://github.com/tacit-code/zenoti-mcp-server.git
cd zenoti-mcp-server
npm install
npm run build

Configuration

Claude Desktop / Claude Code config:

{
  "mcpServers": {
    "zenoti": {
      "command": "npx",
      "args": ["-y", "zenoti-mcp-server"],
      "env": {
        "ZENOTI_API_KEY": "your_api_key",
        "ZENOTI_CENTER_ID": "your_center_guid"
      }
    }
  }
}

The server starts even without credentials and returns a readable configuration error on each tool call, so a misconfigured client shows the problem instead of crash-looping.

Tools

Catalog

Guests

Appointments

Booking flow

1. zenoti-availability-slots — creates a booking draft for guest+service+date and returns booking_id plus open slots
2. zenoti-booking-reserve — hold a chosen slot (holds are short-lived)
3. zenoti-booking-confirm — finalize; the response includes the invoice

Invoices

Reports

Forms & feedback

Zenoti API gotchas this server handles

• Three identifiers in one workflow. Check-in/no-show/feedback take appointment_group_id; details/progress/forms take appointment_id; cancel and payments take invoice_id. Tool schemas name the exact one required.
• Guest updates replace the whole object. The update tool fetches the current profile and merges your changes; a naive partial PUT would erase fields.
• Rate limit is 60 calls/minute (org-wide). The HTTP layer honors Retry-After on 429. Reads also retry on 5xx/network failures/timeouts; writes retry only on 429 (a rate-limit rejection was never processed, so retrying can't duplicate a booking — any other write failure surfaces immediately).
• Pagination caps at size=100; larger values are rejected by Zenoti with a 422.
• Webhooks have no subscription API — they are configured in the Zenoti web UI (Admin → Webhooks) and require the Zenoti API package. This server intentionally has no webhook tool.
• Undocumented numeric enums. Gender and appointment-progress codes aren't in Zenoti's public docs; the mappings used here are documented in the tool descriptions and can be overridden (progress_code).

Development

npm run dev        # run from source (tsx)
npm test           # unit + end-to-end tests (mock Zenoti API; no real credentials needed)
npm run typecheck  # tsc --noEmit
npm run build      # emit dist/

The e2e suite spawns the real server over stdio via the MCP SDK client and asserts every tool's method, path, query, and body against Zenoti's documented endpoints.

License

MIT