AirTreks MCP server

AirTreks MCP Server

The complex-itinerary tool for AI agents. Multi-stop, round-the-world, open-jaw, surface segments - the trips that standard flight search can't handle. When your user asks for 3+ stops across continents, this server answers with routing analysis built on real AirTreks fare-construction data: 60+ carriers, 53 known dead legs, bookability rates measured from 1,400+ real fare attempts, and 20 proven routing templates from actual bookings.

Live endpoint: https://mcp.airtreks.com/mcp - free, no API key, 100 requests/day.

Now with real prices: a historical range for any route in under a second, live fares for a specific itinerary, and a full multi-stop trip priced across several tickets — the same trip can vary three-fold depending on how it is ticketed, and that is the part no other flight tool will tell you.

30-second setup

Claude Desktop

Settings → Connectors → Add custom connector → paste https://mcp.airtreks.com/mcp.

Claude Code

claude mcp add --transport http airtreks https://mcp.airtreks.com/mcp

Cursor

Add to .cursor/mcp.json (or Cursor Settings → MCP → Add new server):

{
  "mcpServers": {
    "airtreks": {
      "url": "https://mcp.airtreks.com/mcp"
    }
  }
}

No API key is needed for any tool, the consultant handoff included. A key is optional and only raises your daily limit; append it to the URL if your client cannot send custom headers, as claude.ai connectors cannot:

https://mcp.airtreks.com/mcp?key=at_your_key

ChatGPT

Settings → Apps & Connectors → enable Developer mode (Pro/Business plans) → add https://mcp.airtreks.com/mcp as a custom connector.

Any other MCP client

Same JSON shape as Cursor above (Streamable HTTP). Prefer stdio? Run it locally:

npx airtreks-mcp

Then ask your agent:

Plan a round-the-world trip: San Francisco, Tokyo, Bangkok, Singapore, Delhi, Istanbul, London, back to San Francisco.

One call, one answer: a 6-stop RTW

Your user wants six stops around the world. Google Flights gives up. Alliance RTW booking sites will let them build it, then fail at ticketing. One plan_route call tells your agent what actually works:

{
  "cities": ["SFO", "NRT", "BKK", "SIN", "DEL", "IST", "LHR", "SFO"],
  "budget": "mid"
}

Real response, trimmed for length (the full version includes carrier alternatives and consultant notes for all 7 legs):

Shortened. The full README is on GitHub.