Developers

Connect your AI assistant

The Everybooking MCP server lets an AI assistant work directly in your account — build quotes, check availability, edit products and pricing, author booking funnels, draft emails, and run your day-to-day operations. It speaks the Model Context Protocol, so any MCP-capable client can use it.

REST API reference Widget SDK reference

Before you start

A Growth plan or higher: MCP access is included on Growth, Pro and Enterprise. On Free or Starter the server answers 402 with an upgrade link.
An API token: Create one under Settings → API Tokens. Give it a name you will recognise later ("Claude on my laptop"), then copy the token from its page — it is the only credential the MCP server accepts.
An MCP-capable client: Claude Code, the Claude desktop app, Cursor, or anything else that speaks MCP over streamable HTTP.

Connect

The endpoint is the same for every account: https://mcp.everybooking.com/mcp

Claude Code

One command, from any directory:

claude mcp add --transport http everybooking https://mcp.everybooking.com/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

Claude desktop app, Cursor, and other clients

Add the server to your client's MCP configuration:

{
  "mcpServers": {
    "everybooking": {
      "type": "http",
      "url": "https://mcp.everybooking.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

3
Check it worked

Ask your assistant to run the connection check. It reports which account it resolved and how many tools it mounted — if the account name is yours, you are connected.
```
Ask: "Run mcp_session_health and tell me which account you are connected to."
```

Choosing a toolset

The full catalog is several hundred tools. Every tool definition costs your assistant context before it does any work, so if you only need one area, mount only that area — the assistant will be faster and cheaper, and it will pick the right tool more often.

Toolset	Covers
bookings	Quotes, invoices, line items, payments, availability.
catalog	Products, categories, pricing, parameters, fees, discounts.
crm	Customers, organizations, deals, leads, email and inbox.
funnels	Booking-funnel authoring, templates, publishing.
operations	Tasks, shifts, staff skills, day sheets, consumables.
ticketing	Ticketed events, seats, admission passes, door scanning.
voice	Voice assistants, call routing, phone numbers.
diagnostics	Health checks, import status, delivery troubleshooting.
full	Everything. The default when you do not choose.

How to choose one

Per connection (simplest): Add ?toolset=bookings,catalog to the URL: https://mcp.everybooking.com/mcp?toolset=bookings,catalog
Per request: Send an X-EB-Toolset header with a comma-separated list.
Baked into the token: A token can carry default toolsets, so every client using it gets the same scope without configuring anything.

Combine sets with commas. An unknown name is rejected with a 400 listing the valid options rather than silently mounting everything.

What you can ask for

A sense of the range, not an exhaustive list — your assistant reads each tool's own documentation once connected.

› “Build a quote for 8 guests, two nights in September, and email it to them.”
› “What is available the weekend of the 17th, and what would it cost?”
› “Add a new cabin, set its per-night pricing, and photograph it.”
› “Build a booking funnel for corporate retreats and publish it.”
› “Which invoices are unpaid and past due?”
› “Write a follow-up email to everyone who quoted last month but never booked.”
› “Set up the deposit rules so groups pay 25% up front.”

Train the AI on your own history

Reply training replays your real past conversations - chat transcripts and inbound texts - through the answering pipeline exactly as it is configured today, captures what the AI would say now, and lets you grade those answers before trusting the auto-reply. Nothing is sent to anyone during training: replays run as synthetic customers, and replayed rows are refused by every outward path.

1. Scan: Ask your assistant (connected with an admin token): "Run run_training_scan." It sorts your history into usable and not-usable conversations, stamps the reason on each, and remembers the verdicts so a re-scan only reads what is new.
2. Replay: "Run run_training_replay." Each usable conversation is answered fresh by the live pipeline - real classification, real quote builder, real reply rules, whatever models your account is set to use - and captured for grading.
3. Grade: "Grade the training replays." The assistant reads each one with get_training_replay (the transcript, a business-knowledge snapshot, and the grading rubric) and files pass / fail / uncertain with record_training_finding.
4. Review: Open Settings -> AI Reply Training (/account/settings/reply-training) to review findings with buttons - each card shows the customer's original message, what the AI would say today, the correct answer where your records carry it, and Approve / Dismiss / Modify buttons. The same queue is list_training_findings with resolve_training_finding. Approving with apply:true also installs the proposed guard - a topic the assistant should hand to a person, or a fact your knowledge base was missing - so the same mistake cannot repeat.
Progress: get_training_status shows the last scan, replay runs in flight, and how many conversations are graded at any moment.

If something goes wrong

401 Bearer token required: The client is not sending the Authorization header. Check it is set on the server entry, and that the value starts with "Bearer ".
401 Invalid API token: The token was revoked, mistyped, or truncated on copy. Mint a fresh one under Settings → API Tokens.
402 Payment required: MCP is not included on your plan. Growth and above include it.
400 Unknown toolset: A toolset name is misspelled. The error lists every valid name.
The assistant cannot see a tool it needs: You are probably on a narrow toolset. Remove the ?toolset= parameter to mount the full catalog, or add the set that owns the tool.