Does NDC Aggregation Need Its Own MCP Layer?

Everyone arguing about AI and airline distribution right now is looking at the wrong altitude.

The public fight has been fought at the GDS level. Who controls the infrastructure. Who owns the interface AI agents will use. Whether MCP becomes the standard everyone builds on, or gets overtaken by something else before it finishes maturing. It is a real fight, and TDN has covered it.

But for the airlines this publication follows, especially across Africa and other emerging markets, the GDS is not the whole picture of how NDC content reaches the trade. Most agents in these markets still rely on traditional GDS today. Katherine Whelan, Airlink’s Chief Commercial Officer, told TDN in May that the majority of agents in Southern Africa still operate through GDS, even as a growing number move to NDC or hybrid models. That fact matters, and it should not be flattened away in the pursuit of a cleaner argument.

But alongside that GDS reliance, a second channel has been quietly building capability of its own. Airlink alone lists six approved NDC aggregator partners on its own distribution page. That is not a sign the aggregator has replaced the GDS. It is a sign that a second, less discussed layer already sits between fragmented airline NDC content and the trade, doing normalization work that the current AI-distribution debate largely ignores. The question worth asking is not whether that layer will eventually replace the GDS. It is whether that layer, quietly, is already positioned to become the infrastructure AI agents actually depend on for NDC content specifically, regardless of how much GDS traffic continues to run in parallel.

NDC’s fragmentation problem is well understood by now. Nobody in this industry needs it re-explained. What is less discussed is how the problem compounds as you move through the distribution chain.

One airline implements one version of NDC, imperfectly, on its own timeline. Katherine Whelan, Airlink’s Chief Commercial Officer, told TDN in May that the airline discovered there were fewer standardised features across the industry than it had originally assumed, and that this discovery alone extended its NDC project well past its original schedule. That is one carrier’s experience with its own implementation.

An aggregator does not deal with one implementation. It deals with dozens, each with its own schema version, its own quirks, its own gaps between what the standard promises and what the airline actually shipped. The aggregator’s entire commercial function is absorbing that variance so an agency, a TMC, or an OTA can plug in once and reach many airlines without rebuilding the integration each time.

Now add an AI agent to that chain, and the query pattern itself changes, not just the volume. A human travel shopper searches a handful of times before booking. An AI agent, optimizing across options, comparing fares, checking availability repeatedly as it reasons through a request, can generate searches at a scale that has nothing to do with how many people are actually flying. Reported look-to-book ratios in the tens of thousands to one, in an AI-driven search environment, are not simply more of the old problem. They are a different problem, one that hits the aggregator layer just as hard as it hits the GDS, arguably harder, since the aggregator is often the point already absorbing the most variance per transaction.

One aggregator has put a specific, public answer to this problem into the market.

The fact: TPConnects launched what it calls the Astra MCP Layer in March 2026, describing it as an orchestration layer built into its Astra NDC platform, designed to normalize NDC schema variation, from version 18.2 through 24.4 and beyond, into a single interface. The stated purpose is that a technology partner or an AI agent integrates once against that normalized layer rather than building a separate connection for every carrier’s schema variant.

TPConnects paired this with a tool it calls ConvertEngine, which the company says reduces AI-driven search volume by 60 percent while holding 97 percent accuracy in production.

That figure has not been independently verified by TDN, and it should not be treated as a confirmed industry benchmark until a carrier or an independent party publishes comparable numbers of their own. What can be said with more confidence is narrower and more interesting. TPConnects has given the industry a concrete shape for what an aggregator-level answer to this problem could look like: normalize the schema first, then attack search volume on top of that normalized foundation, rather than trying to solve fragmentation and volume as separate problems.

Whether this is the only serious work happening at the aggregator layer, or simply the first to be made public, is genuinely unknown, and TDN is not asserting otherwise. Aggregators build ahead of their announcements more often than not. What matters here is not that TPConnects is necessarily ahead of specific named competitors. What matters is that one company has demonstrated, publicly and specifically, that the aggregator layer is a plausible place to build this kind of infrastructure at all.

The obvious question is whether other aggregators will follow TPConnects and build their own MCP layers. That question will answer itself over time, and it is worth watching.

The harder question is different, and it is the one the industry has not really asked yet: does the aggregator become the actual AI distribution infrastructure, with MCP simply serving as the exposed interface sitting on top of it?

Consider what an aggregator that fully embraces this role would actually be doing. It would take fragmented airline NDC content and normalize it. It would orchestrate that normalized content across schema versions and carrier quirks. It would apply search optimization to manage AI-driven query volume. It would handle offer construction and pricing logic across multiple airlines simultaneously. Only then would it expose the result through something like MCP, so an AI agent could query a single coherent interface instead of navigating airline-by-airline fragmentation itself.

In that arrangement, MCP is not the strategic asset. It is the doorway. The strategic asset is everything happening behind the door, the normalization, the orchestration, the intelligence about which content to surface and how to structure it for machine consumption rather than human browsing. Whoever builds that layer well has real influence over how AI agents actually experience airline inventory, regardless of which GDS an airline happens to run on.

This is not a claim that any single aggregator has already secured that position. It has not been demonstrated, and TDN is not asserting it. It is a claim about where the structural opportunity sits, and it is a different question than the one the GDS-level debate has been asking.

This is where the stakes sharpen for the markets TDN covers most closely.

African airline distribution rarely runs through one clean channel. It typically runs through some combination of legacy GDS, direct NDC connections, aggregator relationships, and whatever technology stack a given agency or TMC happens to have built. Whelan was blunt with TDN in May about what that means in practice: there should be no expectation of a quick win, and the current hybrid model, where agencies move between traditional GDS and NDC at their own pace, is likely to persist rather than resolve into something cleaner any time soon.

That layered complexity is exactly the kind of problem an aggregator is built to absorb. It follows that the more capable an aggregator becomes at abstracting that complexity away from the trade, the less pressure sits on any individual African carrier to solve every layer of modern airline retailing on its own. That is a sharper proposition than simply saying Africa is behind on NDC and needs to catch up. It suggests African carriers may not need to fully modernize every layer of their own distribution stack independently, if the infrastructure sitting beneath them does enough of that work on their behalf.

Whether that plays out depends on which aggregators actually build for it, and whether they build it with African market realities in mind rather than retrofitting infrastructure designed for mature markets. That is not guaranteed. But it is a genuinely different way of framing the opportunity than most of the current AI-distribution coverage allows for.

There is a second-order question worth raising here, without letting it take over the argument.

MCP, wherever it gets built, solves access. It lets an AI agent reach data and invoke tools through a standard interface. It does not, on its own, complete a transaction. There is a growing argument across the industry, made by more than one serious travel technology company, that a separate standard is needed for that layer. The Universal Commerce Protocol, built by Google and Shopify and now backed by Amazon, Microsoft, Meta, Stripe, and Salesforce, is designed specifically to handle end-to-end commerce inside conversational AI interfaces, including the parts of a booking that MCP alone does not touch.

If that architecture holds, the stack starts to look something like this: MCP or an equivalent protocol handles access and interface. Aggregation and orchestration, wherever that happens, handles normalization and intelligence. UCP or something like it handles the actual transaction. Payment infrastructure handles settlement. The AI agent sits on top of all of it as the consumer-facing layer.

None of that is settled. The industry has not agreed on how these layers will actually fit together, and anyone claiming certainty about the final architecture is guessing. But the shape of the question changes what an aggregator building for the AI era needs to think about. Building an MCP layer solves today’s access problem. It does not automatically solve tomorrow’s transaction problem, and whoever is thinking two layers ahead has an advantage that will not show up in a press release yet.

Here is a question worth putting directly to the industry, stated as a question rather than a forecast.

Could African markets end up leapfrogging part of the unfinished NDC transition, the way mobile money leapfrogged formal banking infrastructure in several African economies, if AI-native commerce infrastructure arrives and matures before the legacy distribution architecture underneath it is fully modernized?

That is an analogy, not a prediction, and TDN is not asserting that it will happen. Mobile money succeeded because it solved a real, specific gap in financial infrastructure that legacy banking had not filled and was not moving quickly to fill. Whether AI-native distribution infrastructure, built at the aggregator layer or elsewhere, ends up solving a comparably specific gap in African airline retailing is genuinely unknown. But it is the kind of question that deserves to be asked out loud, and so far, nobody covering MCP or UCP publicly has asked it in this context.

Nobody seriously doubts that AI agents will eventually touch airline distribution at scale. That part of the debate is largely settled.

What is not settled is which layer sits between those agents and the underlying complexity of airline retailing. If the answer turns out to be the GDS, then the current public fight over Amadeus, Sabre, and Travelport is exactly the fight that matters, and the industry is watching the right battlefield.

But if the answer increasingly turns out to be the aggregator, a layer that already normalizes a meaningful and growing slice of NDC content for the trade even while most agents still work primarily through GDS, then a significant part of the industry’s attention may be pointed at the wrong altitude. And if that aggregator layer eventually combines normalization, orchestration, AI access, and elements of commerce and settlement into a single stack, today’s arguments about MCP will look, in hindsight, like the opening chapter of a much larger story rather than the story itself.

TPConnects has shown one version of what that opening chapter could look like. Whether anyone else writes the next one, and whether it gets written with African and emerging market realities built in from the start rather than added later, is the question this publication intends to keep asking.

Travel Distribution News covers airline distribution, NDC, GDS dynamics, travel payments, and travel technology with a focus on Africa and emerging markets.