For hotels · Artificial intelligence
AI for your hotel: guest concierge, back-office copilot and MCP server
Three uses, one principle: the AI works on your hotel's data, within the limits you set.
Your guests' AI concierge
- It answers in French or English, before arrival and during the stay.
- It relies on the hotel guide, filled in by your team, and on today's staff notices, which override everything else.
- Its instructions are never to guess a fact about the hotel: if one is missing, it offers to have the front desk confirm it and can pass the request on to you.
- It uses web search for outside information (restaurants, transport, weather); for the hotel, it sticks to your guide.
- It suggests items from your menu and can add them to the cart; guests pay themselves.
- Health: its instructions are to give no medical advice. In an emergency, it points to 15 or 112 (emergency numbers in France); otherwise, it passes the request on to the front desk.
Your back-office copilot
- Ask in plain words: open orders, revenue over a period, bookings, out-of-stock items.
- It prepares the action (marking a dish out of stock, changing a price, pausing a service, moving an order forward) and shows you what it will change.
- Nothing is applied without your click, with exactly the parameters shown: the model is not allowed to write.
- Open to your team according to their rights: a member only sees what their back-office access already opens to them, and price changes stay with administrators.
- Every action is logged under the name of the person who clicked.
The MCP server: your assistant, connected to your back office
The Model Context Protocol (MCP) is an open standard that lets an AI assistant use outside tools. Stayflex exposes your hotel's back office over MCP: the assistant you connect reads your data and acts within the limits of the key you give it.
- Read access: orders and their details, bookings, revenue over a period, menu, service status, the hotel's order statuses.
- Actions, depending on the key: change an order's status, make an item available or not, change its price, pause or reopen a service.
- Simulation by default: the assistant sees the effect of an action before applying it. A price change of more than 40 percent is refused without explicit confirmation.
- Never available to an agent: refunds and cancellations.
- Keys are created and revoked by a hotel administrator (Settings › Agent access), who sees each key's number of calls and changes. Every call is logged and the rate is limited per key.
For your technical team
- Endpoint
POST https://stayflexhotel.com/api/agent/mcp- Authentication
Authorization: Bearer <your key>- Transport
- HTTP (Streamable HTTP), stateless: JSON responses, no SSE stream
- Scope
- one hotel per key; scopes read, write:catalog, write:services, write:orders
Reserved for Stayflex partner hotels: the key is created in the hotel's back office.
Frequently asked questions
- Which assistant can we connect?
- Any MCP client that accepts a remote HTTP server with an authorization header: a URL and a key are all it takes.
- Can the AI change a price on its own?
- The copilot, never without your click. An assistant connected through MCP, only with a key that allows it, in simulation by default; a change of more than 40 percent is refused without explicit confirmation.
- Who sees what?
- In the copilot, each member only sees what their back-office rights already open to them; an MCP key only sees what its scopes allow, for a single hotel.
- How do we cut off access?
- By revoking the key in the hotel's settings: it is refused from the very next call.
- Does the AI concierge make up answers?
- Its instructions are to stick to the hotel guide and today's staff notices for anything about the hotel: if the information is missing, it offers to have the front desk confirm it.
Let's talk about your hotel
Fill in the form on the For hotels page: our team gets back to you within 48 business hours. No subscription: a commission only on what is sold, rate shared during our first call.
Talk to us about your hotel