Write a prompt for most artificial intelligence (AI) assistants today asking to research, book and pay for travel, and you’re likely to receive an answer that your assistant cannot do that right now. The issue is that most assistants are still based on generative rather than agentic AI. In most cases, they can give you answers but can’t yet go off and do anything about it … yet.
That is now changing, and model context protocol (MCP) may well prove to be the answer.
MCP is essentially a layer that sits between the large language models (LLMs) that power AI agents and data sources and tools elsewhere on the internet, acting as a standard translator between them.
The protocol was announced by Anthropic, the company behind the Claude AI ecosystem MCP in November 2024, calling it a “standard for connecting AI assistants to the systems where data lives, including content repositories, business tools and development environments.”
MCP is built on a client-server architecture. The LLM is connected to an MCP client. On the other side, the travel supplier or intermediary builds an MCP server that then connects to the data source or booking tool—things like rates, availability, loyalty information and guest profiles. The client and server communicate using MCP’s standardized language.
Previously, companies would have to build a custom API integration to make those connections, costing time and money. MCP seeks to address that problem.
Since Anthropic’s announcement, other major AI companies—OpenAI, Microsoft, Google and Perplexity—have announced their support for the protocol, meaning that it has become the industry standard.
This means that travel companies can expose their data once, clustering all their API connections into a single MCP server, and instantly become available to all AI assistants that use MCP at scale without building one-by-one integrations.
Applying MCP in travel
Many travel companies are taking the leap with MCP. Kiwi.com announced the launch of its MCP server in August 2025, which allows AI agents to search for flights and receive a personalized list of flights in return.
Stanislav Komanec, CTO of Kiwi.com, said, “There is a huge willingness to go and test these kinds of new AI tools. The customer experience will get better and better, and over time, the way will be agentic. The real challenge is understanding how AI and chatbots can navigate within MCPs and what kind of functionalities should be done within the chatbot versus what should be done on the website for the specific company.”
In September, Apaleo claimed first-mover advantage in the property management platform space to launch an MCP server. Others jumping on the MCP train include Expedia and TourRadar.
Apaleo’s VP of business development Florian Montag said, “It’s definitely simpler to develop MCPs when you have a foundation of API-first structure.”
“With MCPs, AI assistants will access data directly from official hotel and travel systems, instead of relying on generic web content. This means users will get verified, real-time information,” said Pablo Delgado, CEO of hotel tech provider Mirai.
Apps in ChatGPT is an absolutely impactful first step to people buying hotel rooms and travel generally in a way that they haven’t before.
Jason Cincotta, Kismet
The importance of MCP is underlined by the announcement in early October of the launch of apps in ChatGPT based on MCP. For the world of travel, it is significant that two of the seven launch partners are Booking.com and Expedia.
“Apps in ChatGPT is an absolutely impactful first step to people buying hotel rooms and travel generally in a way that they haven’t before,” said Jason Cincotta, co-founder of managed AI-commerce cloud Kismet. “The fact that two of the first seven apps were OTAs tells us that consumers desperately want to use this technology to plan trips.”
Cincotta believes that despite first-mover advantage, the OTAs will not necessarily win out in the new MCP world.
“Having played with the apps, I don’t think it’s a differentiated experience from their mobile app. Second, I think there are pretty fundamental reasons why a direct relationship and the information that a hotel can make available to a guest directly will win out in the long run. It’s a wake-up call in terms of saying that this is clearly what consumers want, your guest wants to use AI to plan their trips and they’re already doing it, even with pretty poor tools.”
Apaleo’s Montag believes MCP will drive innovation, and this was evident at a recent showcase with hoteliers in Munich.
“You are able to build a small agent for very easy use cases in less than two minutes with no line of code on top of an agent builder like N8n. Because you can just tell the LLM this is what I want to do, and it understands via the MCP what it needs to build and adapts the code directly. You could build a copilot that could give you specific reservations based on specific queries, guests who checked out and still have an open folio for example. These use cases very easy to build.”
This early MCP rush is likely to become a torrent, which raises the question of whether all MCPs are created equal.
“Each MCP server can expose completely distinct data, tools and capabilities depending on the organization or use case,” said Delgado, “For instance, one MCP might provide access to hotel availability and loyalty data, while another could serve CRM information, guest preferences or operational workflows.”
It is still early days, and the move to MCP-enabled bookings may take longer.
“Booking involves sensitive issues—trust, payment security, data privacy and accountability—so I expect the first MCP-enabled bookings to appear within closed ecosystems, such as loyalty programs or brand apps, where trust and authentication are already in place,” said Delgado.
He added, “In the travel space, there are only a few early implementations, and while they show progress, the value proposition is still weak. I don’t yet see a strong reason to query this data through ChatGPT instead of simply using Google Flights or the brand’s own app or website, which remain faster and more complete. That said, the technology is evolving at incredible speed.”
Potential pitfalls
Hakan Kanar, CTO of Turkish online travel marketplace Wingie Enuygun, pointed out that while MCPs “unlock While MCPs unlock tremendous potential,” they also pose certain challenges.
“Integration is still complex, especially for legacy systems that were never designed to operate on this unified protocol,” he said.
“Data structures must be standardized and service logic redefined to align with MCP requirements. Another critical aspect is how and where data is processed. The choice of which LLM provider to use, whether OpenAI, Anthropic or Google Gemini, matters a lot because each handles data privacy, retention and authorization differently. LLMs can also introduce latency. Standard APIs, for example, can respond in under one second, but with the LLMs, you have reasoning time so choosing the right models and optimizing response times are important.”
MCP will deliver the vision of agentic AI in travel, but we are still taking our first steps.