ALLFLYGHTS now works with AI assistants — meet our MCP server

Today we're launching the ALLFLYGHTS MCP server. It's live at mcp.allflyghts.com, and it lets AI assistants find real flights through our search engine.

If that sentence makes sense to you, you can skip ahead. If not — here's what it means.

A short explanation

For years, flight search has worked the same way. You open a website, type two airports, click search, look at results. That works fine when you know what you're doing.

But more and more, people don't open a flight search website. They ask an AI assistant. "Find me a flight from London to a small town in Hungary." "What's the cheapest way to get to my friend's wedding in rural Ireland next month?"

The problem is that AI assistants haven't had a good way to actually search for flights. They can describe what flights might exist, but they can't look up real routes, real airlines, or real airports the way a person on Skyscanner can.

That's what we've built. Our MCP server is a way for AI assistants — Claude, ChatGPT, Cursor, and others — to talk directly to our flight engine. The AI gets real route options, with real airlines, just like a human user would.

What's an MCP server, exactly

MCP stands for Model Context Protocol. It's a new standard for connecting AI assistants to outside services. Think of it as a universal plug.

Before MCP, if you wanted an AI to use your service, you had to write custom code for every AI platform. Now, you build it once, and every MCP-compatible AI can use it. It's the same idea as USB — one cable, every device.

We were among the first travel companies to ship one.

What our MCP server does

When an AI assistant connects to our MCP server, it gets three tools:

1. Search locations. Find cities, towns, and airports by name. The AI uses this to figure out exactly where the user wants to go.
2. Get city. Look up a specific place by its ID. Helpful when the AI already knows which location it's targeting.
3. Search flights. Find flight routes between one or more legs. Single trips, return trips, complex multi-city itineraries — all handled.

That's it. Three tools, but they're enough for an AI to build a full trip from scratch.

Why this matters

We didn't build a generic flight search. We built one based on places, not airports.

Most flight search engines start with: pick airport A, pick airport B. That works if you live next to a major airport. But most people don't. Most people live somewhere small, somewhere off the main map, with a few airports nearby and no idea which one is best for any given destination.

ALLFLYGHTS searches by location. You can say "I'm in Marfa, Texas" — a town of 1,800 people with no airport of its own — and our engine works out which airports nearby make sense. Then it does the same for your destination. The result is dozens of real options that traditional flight search would miss entirely.

For AI assistants, this is exactly the right shape. AI agents think about places — "the user wants to go to Greece," not "the user wants to fly to ATH." Our engine matches how AI naturally reasons.

We tested it on camera

Before launching, we recorded two live demos with Claude, Anthropic's AI assistant.

• First demo: Claude picks two random places — Marfa, Texas and Sopron, Hungary — and finds 26 real route options. Then it does the same for Sligo, Ireland to Đà Lạt, Vietnam.
  

• Second demo: Claude plans a 5-leg trip across four continents — Reykjavík to Marrakech to Istanbul to Tbilisi to Almaty. The engine returns 26 routes, including two completely direct itineraries.
  

Both demos are unscripted. Claude picked the cities himself, ran real searches, and gave honest evaluations of the results.

Watch them on our YouTube channel.

How to use it

If you build AI applications or use an AI assistant like Claude Desktop or Cursor, you can connect to our MCP server today.

• Server URL: https://mcp.allflyghts.com/mcp
• Authentication: bearer token, free for development use
• Full setup guide: allflyghts.com/developers/mcp

Sign up for a developer account at allflyghts.com/access/sign-in and you'll have your token in a few minutes.

What's open

The MCP wrapper — the code that translates AI requests into calls to our engine — is fully open source on GitHub at github.com/allflyghts/allflyghts-mcp. MIT licensed. Fork it, study it, contribute to it.

The flight engine itself stays proprietary. But the way you connect to it is open, and we welcome contributions.

What's next

We're already working on the pieces that come after search:

• Pricing. Coming soon
• Booking. Coming soon - stay tuned
• More tools. Saved searches, route alerts, alternative-date suggestions.

Where this fits

ALLFLYGHTS is an independent aviation intelligence platform — independent by design, powered by science, and built on clarity and fairness. We don't take placement money from airlines. Our search ranks by what's best for the user, full stop.

The MCP server is one piece of that vision. As more travel happens through AI agents, the layer that connects them to real flight data has to be neutral and trustworthy. That's the layer we're building.

Try it. Tell us what's broken. Help us make it better.

— The ALLFLYGHTS team