The MCP wave in travel is the most compelling infrastructure shift I've seen in years, precisely because it solves the right problem: access.
The travel technology stack has always been a fortress. Decades of proprietary protocols, arcane fare-filing formats, and GDS middleware that requires a PhD in EDIFACT to navigate. Building a travel application meant hiring specialists who understood not just software, but the peculiar language of PNRs, ticket exchanges, and interline agreements. The barrier to entry wasn't technical complexity alone. It was institutional knowledge, and it was very high.
That barrier just got a lot lower.
Model Context Protocol is the open standard Anthropic released in late 2024 and donated to the Linux Foundation's Agentic AI Foundation a year later. It gives AI agents a universal way to connect to external data sources and tools, discover what's available, understand the schema, and start working, without a developer hand-wiring each integration. MCP crossed from interesting experiment to industry default somewhere around late 2025. It now has over 97 million monthly SDK downloads, more than 10,000 active servers, and first-class support from every major AI platform. OpenAI, Google, Microsoft, AWS, and Block are all founding members of the foundation governing it.
And now it's in travel.
Sabre built a proprietary MCP server on top of their existing API stack and launched it in September 2025, claiming first-mover status. The framing is straightforward: the MCP layer acts as a universal translator, making Sabre's complex travel APIs understandable to any AI agent. Flights, hotels, post-booking servicing, all of it accessible through a standardized protocol that an agent can discover and use dynamically. What's worth noting is that Sabre's MCP connects developers to Sabre's own API, which today surfaces primarily traditional EDIFACT-published fare content, their existing GDS inventory across 420-plus carriers, along with low-cost carrier and lodging content.
Sabre's not the only game in town. Amadeus content is available through multiple community-built MCP servers. Hotels are getting their own MCP endpoints through platforms like Agentic Hospitality's TravelOS, which lets properties show up natively inside AI agents without an intermediary. Many more will come.
And MCP is also not the only pattern in play. The agentic AI stack has two major integration approaches developing in parallel. MCP handles external connectivity: it's how an agent plugs into a system, discovers its capabilities, and calls its APIs. But there's also the "skills" layer, higher-level workflow packages that encode domain logic, quality checks, and multi-step instructions for how to actually get a job done. Skills define what to do. MCP provides the pipes. Tools execute the individual actions. The agents that will matter in travel will use all three, and the best ones already do.
Here's where it gets interesting and, I think, a little dangerous. This stack lowers the barrier to building travel applications so dramatically that we're about to see an explosion of "vibe-coded" travel apps, software built by people describing what they want in plain language while an AI writes the code. Sabre's own developer platform promises production-ready travel features in days, not months. One developer recently documented building an agentic travel planner, front-end and back-end, in a month of spare-time hacking. The same project would have required a team and a six-figure budget two years ago.
Some of these apps will be genuinely useful. They'll come from people who have spent years in the industry, who know exactly which workflows are broken and which edge cases kill you, and who finally have the tools to fix them without hiring a development team. The travel veteran who has been sketching solutions on napkins for a decade can suddenly ship one. That person is dangerous in the best possible way.
But a lot of them will be terrible. When you lower the barrier to building something, you also lower the barrier to building something badly. Travel is a domain where a booking error can strand someone in the wrong city, where fare rules carry legal and financial consequences, and where post-booking servicing requires understanding a web of interline agreements that most developers, human or otherwise, have never encountered. The gap between a demo that shops for flights and a product that reliably handles the full lifecycle of a travel transaction is vast, and I don't think most vibe coders are going to cross it.
Shopping is the demo, not the hard part. The hard part is the exchange at 2 AM agency time when a connection half-way around the world cancels and the passenger is on an interline itinerary with three carriers, two fare bases, and a non-refundable ancillary that only one of them recognizes. While MCP may make it trivially easy to connect to the system that handles that, it does not make it trivially easy to handle it correctly.
The MCP wave in travel is the most compelling infrastructure shift I've seen in years, precisely because it solves the right problem: access. The institutional knowledge barrier that kept the travel app ecosystem small and slow is dissolving. What replaces it will be a mix of brilliant tools from people who understand the domain and a flood of shiny demos from people who don't. The market will sort them out, but not before some travelers have a very bad day.
The question is whether the industry builds carefully or just vibe-codes its way into a mess.
Keep building,
-- JW
