AI enters the GDS: can MCP safely service reservations?
Changing a flight for a fee is one of the processes in which the agency can see the true value of AI. However, a misconstrued request from the passenger can end up in a costly change. It is at this boundary between automation and responsibility that Sabre’s current announcement should be observed.
Sabre reported on September 21 that almost 80 users were using its production software. MCP server. Among them are Virgin Australia, Flight Centre and Internova. Particularly noteworthy for Internova is the development of automation in the reservation service. However, it has not been announced how many users are in production or what savings have been achieved.
MCP connects AI with the booking system
MCP, or Model Context Protocol, enables AI agents to access the functions of the booking system through a standardized interface. The agent can thus use the available tools for searching, booking, and servicing, within the assigned authority.
In its announcement, Sabre describes MCP as the next big change after NDC. However, the technical documentation shows that these two approaches complement each other: MCP enables an AI agent to work with tools that process traditional and NDC content.
NDC offer needs to be re-examined
In the update of September 24th, Sabre added Flight Check to the MCP workflow. Before booking the NDC offer, the price and availability are again checked and updated offer identifiers are returned.
The documentation also warns of a practical limitation: the NDC process requires the AI model to retain more data through conversation. Sabre therefore recommends testing different models and the sizes of their context window, that is, the amount of information they can process during a task. This is a concrete reason why successful demonstrations do not yet prove reliability in everyday work.
Who approves the change and the surcharge?
For travel agencies that use Sabre GDS, it may be best to start with a limited scenario. AI can prepare modification options and calculate the cost, while the agent confirms the execution.
Before broader implementation, payment limits, cancellation approval, a record of completed actions, and the case takeover process need to be agreed upon. When automation stalls, suppliers should be asked to demonstrate compliance with actual tariff rules and confirm the functions available in a specific market.
MCP facilitates the connection of AI with the GDS. The security assessment should therefore be based on what the system is allowed to do, how it verifies the result, and when the work is returned to the person. The trend will certainly go in this direction; other GDSs and aggregators are working in this direction, and there is no doubt that the development will continue. Again, when I think about it, I am not sure how much the aviation industry will save from these developments because, judging by everything, AI tools will also seek to monetize their services.
