TPConnects’ new whitepaper, Adopting Modern Airline Retailing: OOSD, NDC, and MCP, was published on 8 October and makes a simple promise about Model Context Protocol. MCP, it says, is a universal orchestration layer that lets an AI agent integrate once and reach every MCP-enabled airline’s NDC API, while the airline keeps full control over pricing, content presentation and customer data. The paper presents this as the final piece that makes airline retailing omnichannel, including AI channels.
TDN read the paper in full and put three questions on the MCP chapter to TPConnects. Praveen Kumar, co-founder and CTO, responded.
Who runs the layer
Asked whether the MCP layer is operated by the airline, TPConnects or a third party, the answer was that TPConnects develops and manages the layer as the airline’s technology partner, while the airline retains ownership and operational control of the platform.
Both statements can be true in a contract. In practice, they describe a vendor-run service sitting between AI agents and airline NDC APIs. The distinction is between control over airline retailing decisions and control over the infrastructure that carries out AI-agent requests. The airline may retain authority over its offers and customer data while TPConnects operates the intermediary. The answers do not say where the layer is hosted, what traffic data the operator can see, or how an airline could move to a different operator.
Where the pilots stand
Asked whether anything is live today, Praveen described pilot implementations and testing programmes with several airlines, in which AI agents handle the shopping experience while bookings are completed through existing NDC APIs. TPConnects is, in its words, progressively validating MCP-driven transaction flows.
Three things follow from that answer. The response establishes that testing is under way, but does not identify the participating airlines or provide transaction volumes or results. Bookings still run through existing NDC APIs, making MCP, on the description provided, an orchestration layer on top of existing distribution infrastructure rather than a replacement for NDC. The whitepaper, by contrast, says MCP adoption is gaining momentum in 2026 and urges airlines to implement now to secure an early advantage. The only named airline results in the paper, from Air Cairo and Egyptair, concern NDC distribution, not MCP.
The compute cost question
The whitepaper itself raises cost as a risk. It warns that as AI and bot-driven traffic grows, the cloud compute behind real-time offer generation, pricing and AI orchestration can scale unpredictably and erode margins.
Praveen’s answer reframes the issue. Airlines should treat AI agent traffic mainly as a personalisation channel for known, authenticated customers such as frequent flyers, who are expected to reach AI assistants through airline credentials or social sign-on rather than as large volumes of anonymous bots. Guardrails and authentication can limit misuse, and airlines should start with high-value segments where the return is highest. The answer accepts that transaction volumes and compute costs will rise where LLMs personalise offers, points to the scalable consumption pricing of cloud providers, and says airlines can connect their preferred LLM provider, such as Claude or ChatGPT, through the MCP framework. Behind the scenes, it adds, MCP requests remain standard shopping and booking API transactions.
That is a coherent strategy for a loyalty programme. The answer addresses how airlines might limit misuse and prioritise higher-value customers. It does not quantify the cost of agent-driven shopping, say who bears that cost, or explain how the economics change when an AI assistant runs repeated searches before a customer books. That leaves the whitepaper’s margin concern unresolved. It also pulls against the paper’s own pitch, which lists expanded distribution to ChatGPT, specialised travel agents and enterprise assistants as a headline benefit. Reaching third-party agents and focusing on known customers are not necessarily contradictory, but they point in different directions.
What airlines should take from it
MCP may well become the practical way for AI agents to reach airline NDC content, and the integration argument is sound: one connection is easier to build and maintain than hundreds. TPConnects’ answers show a model that is still taking shape, with airline ownership, vendor operation and pilots under way. Control, in this model, depends on the architecture, contractual safeguards and operational arrangements around the protocol, not on MCP itself. Airlines weighing MCP will want clarity on who runs the layer, what that operator can see, who bears the cost of agent traffic, and how pilots progress toward production. Those details will decide whether MCP delivers the control the whitepaper describes.
