Why Your Hotel Needs an MCP Server, and How to Do It for Free (or Not Overpay)

Picture a guest at home, two months before their trip. It's 11pm. They open ChatGPT and type: "Find me a quiet hotel in Seattle with a bathtub and a great view, around $400 a night, for the weekend of June 12."

ChatGPT thinks for a second and recommends three hotels. Yours is not one of them.

This is happening right now, every day, at thousands of hotels. Not because your hotel is worse than the ones that got recommended. Because ChatGPT couldn't read your hotel the way it could read theirs.

The fix is small, mostly free, and most hoteliers have never heard of it. Worse, the vendors who have heard of it are starting to charge five figures a year to install it, which is roughly the same as charging you a thousand dollars to put up a sign that says "We're open."

This post is about closing that gap before someone sells you a more expensive version of it.

The new front desk

For thirty years, the front desk of a hotel has been three things: the lobby, the website, and Google. A guest could call, walk in, or search. Whatever they got back, a person at your hotel had eventually written.

In the last eighteen months, a fourth front desk appeared, and it is the only one a lot of younger travelers use. It's an AI assistant. They ask Claude, ChatGPT, Perplexity, or whichever assistant their phone defaults to, and the assistant answers.

When the assistant answers, it's pulling from whatever it can find. Your marketing copy. Old OTA listings. TripAdvisor reviews from 2019. Anything it can scrape. That's the same starting point your hotel has on Google. The difference is the assistant doesn't show ten blue links. It picks one or two hotels and that's the answer.

If your hotel isn't easy for the assistant to understand, it's not in the answer.

What's actually missing

The assistant has plenty of information about your hotel. What it doesn't have is a way to ask your hotel a question.

It can read your website. It can't ask "do you have a quiet room with a bathtub on a high floor for these dates?" There is no place to send that question. Even if there were, your website wouldn't know how to answer it back in a form the assistant can use.

What's needed is a small piece of software, hosted by your hotel or your tech provider, that AI assistants can talk to directly. Think of it as a dedicated phone line for AI. The assistant calls, your hotel answers, and the answer is precise: yes, room 1108, 11th floor, water view, has a soaking tub, $389 a night, available those dates.

That phone line has a technical name: an MCP server. The letters stand for Model Context Protocol. The protocol is the dialing convention all the major assistants now agree on. Claude speaks it. OpenAI's Agents SDK speaks it. The list is growing every month.

That's the whole concept. A standard way for AI assistants to call your hotel and get a real answer.

Why this matters for your bottom line

Three things follow from having that phone line.

You get recommended more. When an assistant can ask your hotel a precise question and get a precise answer, your hotel is far more likely to be the one it picks. Assistants prefer specific over vague, the way a guest does.

The bookings that come in are direct. A guest who got your hotel recommended by ChatGPT, with a specific room and a specific quote, has no reason to go through Booking.com. You keep the 18% commission.

You stop competing with hotels twice your size. The big chains spend millions on SEO. They spend almost nothing on AI assistant visibility, because it's new and nobody has built the playbook yet. Right now, a 60-room independent boutique with an MCP server and good data is more visible to an AI assistant than a 600-room Marriott without one.

The window where that's true won't last more than a year or two. The hotels that ship now will own the shelf in 2027.

What it actually costs

If a vendor has pitched you on this and they're asking for $500 to $5,000 a month, here is what they're not telling you.

The protocol is open. There is no license fee.

The software is open source. Anthropic released it. Independent developers have released starter templates. There is a working server in a half-dozen programming languages, free, on GitHub.

The hosting costs about $20 a month on the same kind of server that runs your website. If your PMS provider is competent, they can run it for you at no marginal cost. If your website is on Wix or Squarespace, you'd pay a cloud provider directly, and it's still about $20.

The setup is a one-day job for a competent developer. We've done it for hotels in three hours.

What a vendor adds, if they're honest, is two things: keeping the server up to date as the protocol evolves, and making sure it's connected to your PMS so the answers are real. The first is genuinely worth paying for in a small way. The second is worth paying for once, not monthly.

If a vendor is quoting you a recurring price that looks like "AI SEO consulting," ask them to itemize what is software, what is hosting, and what is human labor per month. Most of the time, the answer makes the bill collapse.

What an AI assistant should be able to ask your hotel

A useful MCP server for a hotel can answer questions like:

• Do you have a room available for these specific dates?
• What's the quietest room you have for those dates?
• Which rooms are pet friendly, have a real bathtub, or face the water?
• What floor is room 1108 on, and what's the view?
• What's your policy on early check-in, on pets, on parking?
• Can you hold this room for fifteen minutes while the guest decides?

That last one is the interesting one. A good MCP server isn't just read-only. It lets the assistant act on the guest's behalf in small, safe ways: hold a room, send a quote by email, note a preference on the reservation. Each of those is one less reason for the guest to go to an OTA.

You don't need all of those on day one. Start with availability and room details. Add the rest later.

What we built, in plain terms

Roomza runs an MCP server for every hotel on our platform. It answers about a dozen of the questions above using the room-by-room data we already keep on each property. When a guest asks Claude about a hotel we cover, Claude can call us directly and get a real answer. ChatGPT and Perplexity will be able to do the same as their agent products mature.

It cost us nothing to add. We already had the data. We added a small piece of software that exposes it in the format the assistants expect.

If you're a hotel and you want yours running this month, we'll do it for you for the cost of an hour of our time. Not because we're being generous, but because the actual work is about that long.

What to ask on Monday

If you remember three sentences from this post, make them these.

Ask your PMS provider whether they offer an MCP server. If they say yes, ask what it exposes and what it costs. If they say no, ask when. If they say "what's MCP," forward them this post.

Ask your website company whether your hotel can be queried by AI assistants. If the answer involves the word "schema," that's a start but not the whole answer. Push for the phone line, not just the brochure.

If a vendor is selling you "AI booking optimization" or "AI SEO" for more than a few hundred dollars a month, ask them to itemize. What is software you could buy for $20? What is one-time setup? What is ongoing work? If the bill doesn't shrink under questioning, the vendor is selling you a wrapper around something free.

The shift

For thirty years, hotels learned to compete for the top of a Google results page. The skill set was SEO, paid search, OTA management, metasearch. The cost was high, the margins were lower every year, and the rules changed every time Google changed their mind.

The next thirty years will be different in one specific way. Travelers will ask AI assistants for a hotel and the assistants will answer. The hotels that win will be the ones an assistant can actually read.

The good news is that becoming readable is small, cheap, and largely under your control. The hotels that figure that out in 2026 will be the ones still in business when the OTA commission squeeze tightens further. The ones who don't, or who let a vendor turn a $20 task into a $20,000 contract, will be the ones an assistant has never heard of.

It's a small piece of software. Don't let anyone tell you it's anything more than that.