The era of AI agents that can actually do things — not just chat about them — is here. And one of the most compelling use cases is travel booking. An AI assistant that can search flights, compare options, select seats, add baggage, and confirm a booking — all through natural language conversation.

The missing piece until recently was a standard protocol for AI agents to interact with external tools. That's exactly what MCP (Model Context Protocol) provides. And RunRelay is one of the first travel platforms to ship a production MCP server.

What is MCP?

Model Context Protocol, developed by Anthropic, is an open standard that defines how AI models interact with external tools and data sources. Think of it as a USB-C for AI — a universal connector that lets any compatible AI model use any compatible tool.

Before MCP, every AI integration was bespoke. If you wanted Claude to search flights, you'd write custom function definitions, handle the API calls in your application layer, and parse the results back to the model. Every provider, every tool, every integration was different.

MCP standardizes this. An MCP server exposes a set of tools with defined schemas. An MCP client (the AI agent) discovers these tools, understands their parameters, and calls them as needed. The protocol handles tool discovery, invocation, and result formatting.

RunRelay's MCP Server: 9 Tools for Travel

RunRelay exposes a production MCP server with 9 tools that cover the complete travel booking lifecycle:

  1. search_flights — Search for flights by origin, destination, dates, passengers, and cabin class. Returns normalized results from GDS, NDC, and LCC sources.
  2. get_flight_details — Get detailed information about a specific flight offer, including fare rules, baggage allowance, and cancellation policy.
  3. get_seat_map — Retrieve the seat map for a specific flight, with availability, pricing, and seat characteristics (window, aisle, extra legroom).
  4. search_ancillaries — Search available ancillary services: extra baggage, priority boarding, lounge access, travel insurance, eSIM data plans.
  5. create_booking — Create a flight booking with passenger details, selected seats, and ancillary services. Supports both synchronous (GDS/NDC) and asynchronous (LCC) confirmation.
  6. get_booking — Retrieve booking details by PNR or booking reference, including current status, e-ticket numbers, and itinerary.
  7. modify_booking — Request changes to an existing booking: date changes, name corrections, ancillary additions.
  8. cancel_booking — Initiate cancellation and refund processing for an existing booking.
  9. search_hotels — Search for hotels by location, dates, and preferences. Returns availability and pricing from aggregated sources.

Each tool has a fully typed JSON schema with descriptions, required parameters, and example values — everything an AI model needs to use them correctly.

How It Works in Practice

Here's what a conversation with an MCP-connected AI agent looks like:

User: "I need a flight from Berlin to Barcelona next Friday, returning Sunday. Just me, economy is fine."

The AI agent calls search_flights with the extracted parameters:

{
  "tool": "search_flights",
  "params": {
    "origin": "BER",
    "destination": "BCN",
    "departure_date": "2026-02-27",
    "return_date": "2026-03-01",
    "passengers": { "adults": 1 },
    "cabin_class": "economy"
  }
}

RunRelay returns results from across all distribution channels — Lufthansa via NDC, Vueling via GDS, Ryanair via direct/human-assisted. The AI agent presents the options:

Agent: "I found 12 options. The cheapest is Ryanair at €47 roundtrip (basic, no baggage). EasyJet is €62 with a carry-on included. Vueling is €89 with checked bag. Want me to show more details on any of these?"

The user selects Ryanair. The agent calls get_seat_map to show available seats, search_ancillaries to offer baggage and priority boarding, and then create_booking with all selections.

Since Ryanair is an LCC, the booking enters async mode. The agent tells the user: "Booking submitted! You'll receive confirmation within 10-15 minutes." RunRelay's human operator completes the booking on Ryanair's website, and a webhook fires with the confirmed PNR and e-ticket.

Async Booking Mode for LCC

This is a critical capability that differentiates RunRelay from any other travel MCP server. When the selected flight is on a carrier that doesn't support API-based booking (most LCCs), the create_booking tool returns immediately with a status: "pending" response and a booking reference.

{
  "booking_ref": "RR-2026-BER-BCN-4721",
  "status": "pending",
  "estimated_confirmation": "10-15 minutes",
  "webhook_url": "configured via API key settings",
  "message": "Booking is being processed by our team."
}

The AI agent can then:

This async model means AI agents can book any airline in the world — not just the ones with APIs. The agent doesn't need to know or care which distribution channel serves which carrier. It's all abstracted behind the same MCP tool interface.

Integration: Claude, GPT, and Beyond

RunRelay's MCP server works with any MCP-compatible client. As of February 2026, the primary consumers are:

Setting up the connection is straightforward:

{
  "mcpServers": {
    "runrelay": {
      "url": "https://mcp.runrelay.io",
      "auth": {
        "type": "bearer",
        "token": "your_api_key"
      }
    }
  }
}

Once connected, the AI model automatically discovers all 9 tools and can use them in conversation. No additional code required.

Why MCP Matters for Travel

Travel is one of the most complex consumer transactions. It involves real-time pricing, inventory management, regulatory compliance, multi-party coordination, and high stakes (a wrong booking costs real money). These are exactly the characteristics that make it ideal for AI agent delegation — but only if the agent has access to reliable, comprehensive tools.

MCP provides the standard. RunRelay provides the tools. Together, they enable a new paradigm: AI agents that don't just recommend flights — they book them.

The future of travel booking isn't a better search engine. It's an AI agent that knows your preferences, finds the best options, and handles everything — including the parts that need a human operator.

For developers building AI-powered travel products, RunRelay's MCP server eliminates months of integration work. You don't need GDS contracts, airline agreements, or payment processing infrastructure. You connect to one MCP server, and your agent can book flights on any airline in the world.

Getting Started

RunRelay's MCP server is available in production today. To get started:

  1. Request API access at runrelay.io — you'll receive an API key and sandbox credentials
  2. Connect your MCP client using the configuration above
  3. Test in sandbox — all tools work with simulated inventory, no real bookings
  4. Go live — switch to production credentials when ready

Full documentation, tool schemas, and example conversations are available in the developer portal.