Expedia's B2A MCP server lets agents book hotels — but the 'preferred' flag lives one layer above the protocol

Expedia opened a B2A — business-to-agent — MCP server. AI agents can now query and book hotels through it directly, no human clicking through a screen. If you came here from a search, the short version is: a machine can now talk to Expedia’s inventory the way a travel agent used to talk to a GDS. That part works.

I spent years in hotel B2B distribution. A lot of that time went into one unglamorous thing: making the preferred-rate flag move correctly across channels. So when I read the B2A announcement, that’s the part my eye went to. Not “can the agent book.” It can. The question is what the agent sees when it looks.

What “preferred” actually is

When a corporate buyer signs a hotel program, they negotiate rates with a set of hotels. Those become the preferred properties. The whole point is that when a traveler searches, the company’s negotiated hotels surface first, and the company’s contract gets honored.

Here’s the thing people outside distribution miss. “Preferred” is not a property of the room. It’s not in the rate. It’s not in the inventory record. It’s a relationship that lives one layer above all of that — a business agreement between a buyer, a distributor, and a hotel. For years, the systems that surfaced it correctly were built by hand, channel by channel, and they broke all the time. I’ve sat in the rooms where we argued about why a preferred hotel dropped to page three on one point of sale and not another.

So the inventory record an agent reads through MCP is honest. It just doesn’t carry the one signal that decides which of forty identical-looking options the traveler is actually supposed to book.

The protocol can carry the signal. The architecture decides if it does

I want to be precise, because this gets reported wrong. MCP is not broken. The protocol can absolutely carry a “this is preferred for this buyer” field. That’s not a technical limitation.

The open question is how B2A implementations choose to expose it. Does the agent get told which inventory is preferred for the specific company whose travel it’s booking? Or does it get a clean, neutral list where every Marriott looks like every other Marriott and the negotiated relationship is invisible?

If it’s the second one, two things happen, and they hit both sides of the table.

The buyer loses contract enforcement. They negotiated rates assuming their travelers would be steered toward those hotels. An agent that can’t see “preferred” books the cheapest or the closest, and the program quietly stops mattering. Compliance numbers that took a procurement team a year to build start sliding, and nobody can point at the moment it happened.

The distributor loses differentiation. If every option looks the same to the agent, price is the only axis left. The distributor who spent years building relationships, content, and service has no way to say “this one deserves the agent’s attention, and here’s why.” They get flattened into a commodity feed.

We solved this once. For humans

This is the part that keeps me up a little. We already did this work. The entire apparatus of modern travel distribution — the rate codes, the negotiated content, the preferred flags, the reasons one option ranks above another — was built so that a human looking at a screen would see the right thing first. Decades of plumbing, all aimed at one species: people with eyes, reading a results page.

Agents don’t read a results page. They read a structured response and they optimize on whatever fields are in it. If the field that encodes the business relationship isn’t there, the relationship doesn’t exist as far as the agent is concerned. It’s not that the agent ignores it. It can’t see it.

So the work has to be done again. Not the protocol work — that’s the easy part now. The hard part is the same hard part it always was: deciding which signals deserve to travel up the stack to the thing making the decision, and making sure they actually arrive. The smartest distributors will treat this as the moment to lead, not a compliance chore to bolt on later.

What I’m watching for

The interesting B2A implementations won’t be the ones that just expose inventory. Anyone can dump a room list into a structured response. The ones that matter will carry the buyer’s context with the request — who is this for, what program are they on, what’s preferred for them — and answer with inventory that’s already shaped by that context before the agent ever sees it.

That’s the layer I think about constantly. When a buyer’s agent books through us, the preferred relationship has to be present in what the agent reads, not living in a contract PDF nobody parsed. Otherwise we’ve automated the booking and lost the reason anyone built a program in the first place.

A direct API to inventory is necessary. It was never the hard part. The hard part is the layer above it, and we’re about to find out which players remember that.

This is the layer we work on building Bitravel — making the buyer’s program and preferred relationships visible to the agent at the moment it books, not after. Book a 30-minute call for a live look, no deck.