Docs›AI & automation

Connect Mangobeds to AI tools with MCP

9 min read

Your bookings, availability and guests shouldn’t be locked behind a dashboard. The MCP app connects Mangobeds to the AI tools with MCP support you already use, from Claude and ChatGPT to Cursor, VS Code and Codex, so you can ask about your property in plain language and get answers straight from your live data.

Every tool uses the same server address, and this page is the map: what MCP is, the two ways a tool can connect, what your assistant can read and change, and a link to the step-by-step guide for whichever tool you use.

Why connect Mangobeds to AI tools with MCP?

Your property data is most useful when you can just ask for it. Instead of clicking through reports, you open the AI assistant you already use and ask a question. Mangobeds answers from your live bookings, availability and accommodations.

  • Ask your data anything: questions like “Which guests check in this weekend?”, “How is occupancy next month?” or “Find the booking for Andrés”, answered from your live Mangobeds data.
  • Use the AI tool you already have: Claude in the browser and on your phone, ChatGPT, Claude Code, Claude Desktop, Cursor, VS Code, Codex and Antigravity all connect to the same server.
  • You decide what it may do: when you sign in, you choose which organization the assistant can see and whether it may only look or also make changes.
  • No code required: for Claude in the browser and ChatGPT there is nothing to copy at all. You paste one address, sign in, and approve.

Table of contents

What is MCP?

MCP, the Model Context Protocol, is an open standard introduced by Anthropic that lets AI assistants connect to external tools and data in a consistent way. Think of it as a universal adapter between an AI assistant and the apps you run your business on. It is now supported across the major AI assistants, so one connection works in many tools.

Mangobeds runs an MCP server for your account. Once you connect it, your AI assistant can read your data on demand, and make the changes you have allowed.

The two ways to connect

Every tool uses the same server address. What differs is how it proves who you are, and that is not a choice you make: it depends on what your tool is able to send.

  • Sign in and approve, used by Claude in your browser and on your phone, and by ChatGPT. You paste the server address, Mangobeds asks you to sign in, you pick the organization and tick what the assistant may do, and you are sent back. There is no API key to paste, and you should not try to give one. These tools cannot send a key at all, which is exactly why this route exists.
  • Your API key, used by Claude Code, Claude Desktop, Cursor, VS Code, Codex and Antigravity. These run on your own computer and can send a key with each request, so you paste your key into their configuration once.

The MCP page in Mangobeds tells you which is which: each tool is shown as a card that says either “Signs in, no key” or “Uses your API key”, and picking one shows the exact steps for it.

The difference matters beyond convenience. A key belongs to your whole organization, so everyone using it looks the same and turning it off turns off every integration at once. Signing in belongs to you: your connection can be cut on its own, your name appears against anything the assistant changes, and if you ever leave the organization your connections stop working automatically.

Pick your AI tool

Each tool has its own guide with the exact steps, the file or command it needs, and the things that catch people out.

Sign in, no key needed

Uses your API key

Other MCP-compatible tools, Windsurf among them, follow the same shape: point them at the MCP server URL from Apps > MCP and pass your key in the X-API-Key header.

What you can ask, and what it can change

Your assistant can find its own way around: it can list the properties in your account, so you never have to look up an ID or paste anything from the dashboard. From there you can ask about:

  • Bookings: “Show the bookings checking in this week”, “Find the reservation for Maria Lopez” or “What was the last booking we took?”
  • Availability: “Which rooms are free next weekend?” or “What is my occupancy in August?”
  • Blocked rooms: “Why is room 4 not bookable next Tuesday?” It can tell the difference between a room that is sold and a room you have blocked, and show the note you left.
  • Prices: “What does the double room cost on 20 September?” or “Show me the seasonal rates on this room type.”
  • Guests: “How many bookings has this guest had?” or “What did they put in the custom fields?”
  • How Mangobeds works: it can read these help guides, so “how do I set up seasonal rates?” is answered alongside what your own rates actually say.

If you allow it, the assistant can also make changes: correct a guest’s details, add an extra guest to a booking, and send one of your saved messages to a guest. That last one delivers a real email or WhatsApp message, so it is worth being deliberate about. Anything it changes is written into the booking’s activity with your name and the assistant’s, so you can always see what happened and who authorised it.

Using Windows? Read this

On Windows, some desktop tools do not escape spaces inside a command’s arguments. The space in X-API-Key: YOUR_API_KEY splits the value in two, so your key never reaches Mangobeds and the connection fails with an authentication error. Your key is fine, it is the space that breaks it.

The fix is the configuration in the Claude Desktop guide: no space after the colon, and the key passed as an environment variable. That same file works on macOS and Linux, so you can use it everywhere if you prefer one set of instructions.

This does not affect Claude in the browser or ChatGPT, which never send a key in the first place, and it does not affect Cursor, VS Code, Codex or Antigravity, which keep the key in its own field rather than on a command line.

Permissions, and disconnecting

When you connect by signing in, the approval screen asks two things: which organization the assistant may see, and whether it may only look or also make changes. Untick “make changes” and it can read everything and alter nothing. You can connect again later with different permissions if you change your mind.

Every connection made this way is listed under Connected assistants on the MCP page, showing which assistant it is, who signed it in, what it is allowed to do and when it was last used. Disconnect ends that one connection and leaves every other one running, so cutting off one person’s Claude does not break your other integrations.

Connections made with the API key work differently: the key is your organization’s, so deleting it disconnects everything using it at once. Only do that if the key has leaked.

Frequently Asked Questions

Is my data safe?

The assistant only ever reaches the organization you chose, and only does what you allowed. Connections made by signing in appear under Connected assistants and can each be cut on their own; a connection made with the API key stops the moment you delete that key. If you leave an organization, any connection you signed in with stops working straight away.

Can the assistant change my bookings?

It can if you let it. When you connect by signing in you choose whether it may make changes; leave that unticked and it can only read. With permission, it can correct guest details, add a guest to a booking, and send one of your saved messages to a guest. It cannot create or cancel bookings, or take payments. Everything it does change is recorded in the booking’s activity with your name and the assistant’s next to it.

Which AI tools with MCP support work?

Claude in your browser and on your phone, ChatGPT, Claude Code, Claude Desktop, Cursor, VS Code, Codex, Antigravity and other MCP-compatible tools such as Windsurf. There is a guide for each of the eight above, and they all use the same server address. New clients add support regularly.

ChatGPT is asking me for an API key. What do I enter?

Nothing. Choose OAuth instead. ChatGPT cannot send an API key to a connector, so there is no value that would work, and older instructions telling you to add an X-API-Key header were wrong. The same is true of Claude in the browser: leave the OAuth fields empty and click Connect.

It is not connecting on Windows, what do I do?

Almost always it is the space in the API key header, which splits the value so the key never arrives and you see an authentication error. Use the configuration in the Claude Desktop guide, which passes the key as an environment variable and removes the space.

My assistant says it cannot find my property. Why?

Start a new chat. The assistant can list your properties itself, but it only picks up new abilities when a conversation begins, so a chat started before you connected, or before an update on our side, may still be working from the old set. If a new chat does not help, turn the connector off and on again in your AI tool’s settings.

Do I need a developer?

No. For Claude in the browser and ChatGPT there is nothing to copy but one address, and you sign in the same way you sign in to Mangobeds. The tools that use an API key need one file pasted into a settings folder, and every guide gives you that file with your key already in it.

Ready to talk to your property? Go to Apps > MCP, pick your AI tool, and follow the guide for it.