Understanding MCP and its impact on the travel industry

Ira Vouk explains the MCP protocol, its impact on travel payments, the challenges for hotels, OTAs, and airlines, and the future of AI-driven booking.

How would you define MCP for someone in travel payments unfamiliar with the protocol?

Think of the Model Context Protocol (MCP) as the USB-C for AI agents. It was introduced by Anthropic in late 2024 and quickly adopted by OpenAI, Microsoft, and Google in early 2025.

The goal is to replace the fragmented, platform-specific ‘plugin’ model with a universal connection standard. Instead of having to build one custom integration for each AI system, a company (for example, a travel payments provider or booking platform) can build one MCP server. Any AI model that supports MCP can then instantly connect (’mount’) that server and use the tools and data it exposes.

An MCP server acts as a connector layer, exposing functions (tools) such as processPayment, verifyCard, refundTransaction, or getBookingDetails, along with data resources like customer profiles, transaction histories, or payment policies.

The key idea is to have one integration, many AIs. Just as USB-C made hardware connections universal, MCP makes data and functionality universally accessible to AI agents across ecosystems.

What are the main trends driving MCP adoption in travel, and which stakeholders stand to benefit most?

The reasons everyone in travel is suddenly talking about MCP are simple but exciting. Firstly, AI has become the new travel agent. Travellers aren’t typing ‘hotel in Miami’ into Google anymore, they’re telling their AI assistant, ‘find me something cute, clean, not too expensive, and preferably not next to a nightclub.’ And for an AI to answer that without hallucinating, it needs a clean, universal way to talk to hotel systems. MCP is that universal port. 

Another big trend is that the old way of building integrations is… well, let’s call it ‘a museum piece.’ Historically, if you wanted to connect your hotel system to 10 platforms, you built 10 separate integrations, each with its own quirks, drama, and maintenance nightmares. MCP flips that upside down. Build one MCP server, expose your tools once, and suddenly every AI that supports MCP can plug in without you rewriting your tech stack for the seven-millionth time. It’s the closest thing we’ve ever had to a universal adapter in travel tech.

There’s also a very real ‘take back control’ energy in the industry right now. AI-driven search and bookings are a new distribution funnel, and nobody wants to be caught flat-footed while OTAs sprint past them again. Hotels finally see a chance to put their content, availability, rates, and even payment flows directly in front of AI assistants instead of praying that an intermediary will do right by them. MCP is basically the open door back into the conversation. 

And then there’s the rise of agent-to-agent (A2A) automation: AI agents negotiating (potentially), validating, and even booking with each other without a human babysitting the process. MCP is the plumbing that makes this possible. It standardises the way those ‘little robot negotiators’ talk to booking systems and payment systems in real time.

As for who benefits most? Hotels and hotel groups could eventually get a direct, future-proof line into AI ecosystems, meaning no more custom integrations per platform. CRS, PMS, booking engine vendors, and payment processors get to expose their capabilities once and have every AI consume them without extra engineering legwork. Payment companies, especially, get a golden ticket: AI agents can suddenly verify, authorise, and settle transactions through standardised MCP tools instead of duct-taped API chains. 

In short, MCP is taking off because AI is eating the travel distribution funnel, everyone is tired of the fragmentation and the integration drama, and the industry finally sees a path toward being first-class citizens in AI search instead of spectators. The early adopters, suppliers and tech providers who plug in now, will be the ones AI agents actually ‘see’ when this new channel matures.

What are the key limitations or risks for hotels, airlines, and OTAs in this shift?

The biggest limitation in this whole shift is that everyone wants to plug into AI, but nobody actually controls the socket. Hotels, airlines, and OTAs can build beautiful MCP servers all day long, but until AI platforms fully open their app ecosystems, you’re basically standing at the door with your bags packed, waiting for the bouncer to decide if you’re on the list. MCP gives you the key, but someone else still owns the lock.

For hotels, the risk is obvious: if you don’t adapt your data and infrastructure early, AI assistants will default to whoever did. And those early players will become the ‘default recommendations,’ the same way Amazon’s algorithm quietly decides which blender you buy. There’s also the operational side – if your CRS or PMS can barely handle today’s requests without a meltdown, letting AI agents hammer it with real-time queries may turn your system into a crime scene.

Airlines face a special flavour of chaos. They have more rules, restrictions, exceptions, fare classes, sub-fare classes, and conditional surcharges than any other industry on Earth. MCP doesn’t magically simplify the fare universe. An AI agent can query your system perfectly and still get back a funhouse mirror of logic because the underlying data is a maze. If airlines don’t clean up their structures, they’ll end up with AI that follows the rules so literally it refuses to sell half their fares.

OTAs are in a tricky position too. They stand to benefit from MCP, but they’re also staring at the possibility that AI becomes the new meta-distributor. If travellers ask AI directly for hotels and flights, the OTA’s homepage becomes optional. Their advantage was always aggregation and search. AI can do both. So the risk is losing the top of the funnel unless they aggressively adopt MCP and become indispensable to AI agents instead of trying to fight them.

Across all three categories, there’s one universal limitation: data quality. MCP standardises the connection, not the content. If your availability, rates, policies, content, or payment logic are messy today, MCP will simply expose that mess faster and more visibly. Think of it like turning on all the lights in the room – great if you’ve cleaned, not great if you’ve been ‘meaning to tidy up’ for the last decade.

Finally, there’s the strategic risk. The shift to AI as the primary interface puts massive power in the hands of a few platform providers. If you don’t move early and define your place in this ecosystem, someone else will define it for you. And hotels, airlines, and OTAs all know exactly how that story went the last time it happened.

What needs to happen before travellers trust AI to handle their bookings and payments?

Travellers will trust AI with bookings and payments once the process feels as reliable as every other digital transaction in their lives, and honestly, we’re already there. You can book a hotel end-to-end inside Perplexity today. The payment piece isn’t rocket science; digital wallets, secure checkout flows, and PCI-safe handoffs have existed for years.

The real shift isn’t about technology catching up, it’s about travellers realising that AI is just another interface on top of systems that already work. Once people see that an AI booking is no different from tapping ‘Pay with Apple Pay’ on any website, trust becomes a non-issue.

What milestones should we expect for MCP-enabled travel bookings by the end of 2026?

By the end of 2026, we should expect MCP-enabled travel to mature fast, though at very different speeds across AI platforms. Perplexity is already ahead of the pack with its Tripadvisor and Selfbook partnerships, and you can complete a hotel booking end-to-end inside the interface today. Google has publicly committed to the same path: Trip Planning inside Gemini and AI Mode, where the entire transaction happens without leaving the assistant, backed by partnerships with major hotel brands and OTAs. And OpenAI has signalled that its app marketplace will open to travel companies beyond the current OTA duo, which finally gives suppliers and tech providers a seat at the table.

The first wave of direct MCP bookings will be live. Not theoretical. Not demo-only. Actual transactions where an AI agent queries availability, rates, policies, and payment options straight from a supplier’s MCP server and books without going through a legacy intermediary. Hotels will lead that wave because their ARI structures are simpler, but airlines won’t be far behind once fare logic stabilises.

Agent-to-agent (A2A) operations will move from ‘interesting whitepaper’ to ‘we actually use this.’ Think AI agents modifying bookings, issuing refunds, holding inventory, verifying IDs, or rebooking disrupted itineraries without a human sitting between systems. MCP is the plumbing that makes that possible.

And finally, we’ll see the emergence of real trust signals: MCP-verified suppliers, standardised booking confirmations, consistent payment flows, and AI interfaces that show availability, prices, policies, and fees without guessing or scraping. Once the experience feels boringly reliable (as in, ‘sure, just book it through my AI’) the new channel is officially established. 

And that will happen sooner than we think.

This article is part of The Paypers’ Travel Series, which includes contributions on topics spanning emerging trends in travel payments, fraud and security challenges, regulatory and tax impacts, risk management and forex, as well as sustainability in the travel industry. For a complete overview of all the contributions featured, click here. 

About the author

Ira is a hospitality consultant and tech innovator who brings two decades of practical industry experience, predominantly championing the role of revenue management and the use of data and technology to provide hospitality companies with insight to lead their business strategy better.  

She is also a public speaker and educator on revenue and profit optimisation, distribution, and hospitality tech and currently teaches Hospitality Technology at Payne School of Hospitality and Tourism Management at SDSU. 

Ira is also a contributing writer for various hospitality publications and platforms and a guest lecturer at various universities worldwide. Find out more about Ira at http://www.iravouk.com