Cresavo Gateway

Open source · REST v1 · Streamable HTTP MCP

Let AI assistants book salon appointments. The customer approves every one.

Cresavo Gateway is the agent-facing front door to Cresavo salons. Your assistant finds free times and proposes a booking. The customer says yes on their own phone. Only then does the salon book it, exactly as approved.

Demo phase: the salon behind this gateway, “Salong Eko”, is fictional. Use fictional contact details only.

How a request travels An AI assistant proposes a booking through the gateway. The salon asks the customer to approve it on their phone. After approval the assistant calls create_booking through the gateway, and the salon books it. AI assistant Gateway Salon Customer Booked propose forward ask create_booking
The gateway checks the salon's access on every request and never decides on its own.

Principles

Consent built in

Every booking runs on a mandate the customer approved. Change the price, staff or time and it no longer counts.

No database

Only in-memory rate-limit counters keyed by IP. No logs of bodies or credentials. The salon owns the calendar, the data and every decision.

Standard protocols

MCP over Streamable HTTP for agents, REST with OpenAPI 3.1 for everything else. Same operations on both.

Safe to retry

Timed out? Retry with the same idempotency key and get the original booking back, never a duplicate.

A booking, end to end

What your assistant sees over MCP, and what the customer sees on their phone.

  1. CustomerBook me a haircut with Alva on Monday morning.
  2. list_servicesKlippning · 45 min · 650 kr
  3. get_availability09:00 · 10:00 · 11:00 free
  4. create_proposalpending sent to the customer
  5. AssistantI've sent the booking to your phone. Approve it there and I'll finish.
  6. get_proposalapproved mandate received
  7. create_bookingconfirmed Mon 09:00 with Alva
  8. AssistantDone. You're booked for Monday 09:00.

Price or terms changed

The mandate no longer matches: mandate_denied. Nothing is booked.

Slot taken

Someone else booked first: slot_unavailable. Propose a new time.

Approval expired

The customer waited too long: mandate_denied. Start a new proposal.

Connect

Two transports, one contract. Pick whichever your agent speaks.

MCP Streamable HTTP

One JSON-RPC message per request.

https://gateway.cresavo.com/api/v1/mcp

Add it to any MCP client that supports remote servers and custom headers:

{
  "mcpServers": {
    "cresavo": {
      "type": "http",
      "url": "https://gateway.cresavo.com/api/v1/mcp",
      "headers": { "Authorization": "Bearer ${CRESAVO_AGENT_KEY}" }
    }
  }
}

REST OpenAPI 3.1

The same operations as plain HTTP.

https://gateway.cresavo.com/api/v1
curl "https://gateway.cresavo.com/api/v1/services?business=demo" \
  -H "Authorization: Bearer $CRESAVO_AGENT_KEY"

Download the OpenAPI contract. Discovery for agents: /.well-known/usp and /llms.txt.

Credentials. Agent credentials are issued by the salon provider, never by the gateway. Send them in the Authorization header only. Never put them in URLs, prompts or chat. To request demo access, open an issue.

MCP tools

Read tools are safe to call freely. Write tools always end at the customer.

ToolWhat it doesScope
list_servicesServices, prices and staff for a salonservices:read
get_availabilityBookable starts for a service and dateavailability:read
create_proposalPrepare a booking for customer approvalbookings:write
get_proposalStatus of a proposal and, once approved, its mandatebookings:read
create_bookingBook under an approved mandatebookings:write
get_bookingStatus of a booking your client createdbookings:read
cancel_bookingNot available yet: cancellation will need its own customer approvalunavailable

USP-style aliases (usp_services_list, usp_availability_query, usp_bookings_create and more) map to the same native inputs. Full USP conformance is not claimed.

Limits and errors

Per client IP. Hosted assistants share egress IPs, so these caps apply to all their users together; contact the salon provider for higher limits. The salon adds its own limits per agent and per customer.

60/min

requests per IP

5/hour

proposals per IP

5/hour

bookings per IP

HTTP/1.1 429 Too Many Requests
Retry-After: 1800

{"apiVersion":"v1","error":{"code":"rate_limited","message":"Too many requests. Retry later."}}

What the gateway never does

These are tested, not promises. Read the full security model.

  • Invent a booking result. No salon answer means no booking.
  • Reuse an access check across requests.
  • Forward browser sessions, cookies or client IP headers.
  • Store customer data, or log credentials or request bodies. It has no database; only in-memory rate-limit counters keyed by IP.
  • Take payments, export data or bulk delete.

Found a problem? Report it privately.