In travel, we are all excited about the potential of artificial intelligence (AI) agents to provide an improved user experience. Specifically, agentic travel commerce promises a future where AI can understand our intent, propose a highly relevant trip and execute booking and payment.
It’s an exciting vision, informed by one of the most significant technology advances of modern times. As the market develops, it is important to distinguish between easier connectivity and the broader capabilities required to make agentic travel work in practice.
The purpose of this article is to explain the role of model context protocol (MCP) in travel distribution, what it enables and where the wider complexity of travel commerce remains.
What is MCP?
MCP provides a common standard for AI agents to connect and interact with data sources.
A hotel or airline can use an MCP server, for example, to make selected services from its underlying systems and APIs discoverable to an AI agent. MCP is a useful and well-designed protocol, but it does not itself provide intelligence, decision-making, orchestration or automation. The value ultimately depends on the content, business logic and operational capabilities made available through it.
MCP is one part of a broader and rapidly evolving protocol landscape. Commerce-oriented protocols such as Universal Commerce Protocol (UCP) may support additional parts of the shopping and transaction journey. These approaches may ultimately prove complementary rather than interchangeable. Amadeus is working with Google and other industry participants to help develop UCP for lodging.
MCP-based travel solutions are beginning to support live use cases, from conversational flight search to booking and selected servicing workflows. These developments are evidence that standardized agent connectivity can accelerate innovation, and they should be recognized as meaningful progress. However, the existence of a live MCP connection does not demonstrate that the broader requirements of travel commerce have been solved across content, providers, journeys and servicing scenarios.
The question for travel distribution
As MCP makes it easier for travel providers to expose capabilities to AI agents, an important question is whether those agents will increasingly search and book directly with individual providers, reducing the need for aggregation and distribution services. This is a credible scenario to examine, but connectivity alone does not address the wider requirements of travel commerce.
Connectivity is only one part of the answer
It can be relatively straightforward for a provider with an existing API to make capabilities accessible through MCP. But easier connectivity does not remove the underlying complexity of travel commerce. Here’s why:
- Business logic and fulfillment
While an AI agent might be able to complete a simple booking via an MCP connection with an airline, a flight reservation involves complex business logic in areas like ancillary selection, seat maps, baggage rules, regulatory compliance, interline partnerships and dynamic pricing, complex logic that sits within underlying travel technology and systems of record.
MCP can provide a means of connecting to capabilities; it does not itself provide this business logic. Travel purchases must also support servicing requests like ticket changes, adding or removing services, cancellations and refunds. Travel commerce is about far more than the initial search, with an ongoing requirement to reliably execute complex changes.
In other words, an AI assistant may become the interface through which a traveler makes a request, but the underlying transaction still needs to be priced, executed, fulfilled, recorded and serviced. - Cost
If each AI agent were to consult each provider directly via MCP, it would add significant additional infrastructure, processing and cost pressure on both providers and AI companies.
Today, a booking made via Amadeus using traditional EDIFACT connectivity requires an average of 4,000 background transactions, while a New Distribution Capability booking requires closer to 20,000.
Should AI agents begin to query provider systems at scale, this look-to-book ratio would rise to prohibitive levels, with providers incurring significant additional IT costs simply to make their offers available.
Similarly, AI companies operating the large language models (LLMs) that power AI agents would also be exposed to additional costs. LLM infrastructure requires significant computing resources. It simply isn’t efficient to conduct a travel search by sending an AI agent over the web to query hundreds of providers. Instead, aggregation remains vital. Replacing a direct interaction with one provider is very different from replacing the aggregation and comparison capabilities of a distribution platform. - Latency and search quality
Travelers expect fast, accurate search results that display the most competitive pricing options from the whole of the market, with very low latency.
The more providers an AI assistant needs to query individually to provide comprehensive comparison shopping, the greater the cost, complexity and latency involved. Aggregation solves this by providing broad, harmonized content through a common infrastructure.
By contrast, Amadeus answers 3 billion flight search requests every day and is able to return relevant, up-to-date dynamic prices at low latency. This capability is reinforced by our role as a passenger service system provider to a significant number of airlines, giving us close integration with the operational systems where availability, schedules, fares and servicing information are maintained. This improves the accuracy of the content we provide to travel sellers and AI-driven channels.
How MCP fits with travel distribution
MCP will play an important role in agentic travel commerce, and Amadeus supports its adoption as a valuable industry development. However, speed to expose an interface is only one measure of progress. The more demanding test is whether an agent can reliably complete a real journey across search, comparison, booking, payment and servicing, while respecting supplier rules and maintaining an accurate system of record.
Many of the underlying capabilities required to do this already operate through Amadeus APIs and integrations at global scale. Our focus is therefore not simply on adding a protocol layer, but on making our travel content, business logic and execution capabilities accessible to AI-driven channels. This includes our investments in intelligent search, booking and servicing, agent-to-agent communication and emerging commerce protocols. This work involves collaboration with AI companies including Microsoft, Google, Nvidia and Anthropic as the industry works out how agentic commerce can deliver value for travelers and providers. As MCP adoption grows, we will support it alongside the wider capabilities needed to move from AI demonstrations to reliable travel commerce.
For Amadeus, that means enabling access to broad, aggregated travel content, accurate and up-to-date pricing, low-latency search, deeply integrated business logic and advanced servicing capabilities while helping shield provider systems from excessive agentic search volumes.
What would it take to replace aggregation?
What would need to happen for direct agentic commerce to replicate the value aggregation provides today?
That would require several significant developments. Airlines would need to make transactable capabilities broadly available to third-party AI agents, at a scale and on commercial terms that support widespread comparison shopping.
This is not simply a technical decision: providers must also decide which capabilities they want to expose to third-party agents, under what commercial conditions, and at what level of search volume.
AI-driven channels would also need to reliably manage the complexity of multi-airline journeys, including interline ticketing and settlement. And they would need to address what happens after the initial booking, from changes and cancellations to refunds and disruption management, while maintaining the reliable systems of record, accountability and servicing that travel requires.
We do not see these capabilities operating at the breadth, reliability and scale required to replace aggregation today.
The next phase of agentic travel should therefore not be judged by protocol availability alone: The breadth of content available, the number and complexity of transactions completed, response times, servicing completion, exception handling and the level of human intervention required all matter. Early deployments are important, but the long-term benchmark will be whether these capabilities can operate reliably across the full lifecycle of a trip.
The opportunity
The interface through which travelers search and book may change; the underlying complexity of travel commerce does not disappear.
Our opportunity is to make that complexity invisible, connecting new AI interfaces with the content, technology, business logic and execution capabilities that make travel work.
As agentic commerce develops, we believe Amadeus can play an important role in turning AI-powered recommendations into reliable, real-world travel transactions at scale.
About the author...
Sylvain Roy is chief technology officer at
Amadeus.