Do not index
Do not index
There's a new acronym circling the hotel tech world: MCP. Model Context Protocol. Before you move on , this one is worth understanding. Not because it's ready to deploy next week, but because if it gains traction, it changes the structural relationship between hotels, AI agents, and OTAs. That's a meaningful claim, and this post will justify it or tell you where it falls short.
What MCP Actually Is — And What It Isn't
The simplest way to describe MCP is a switchboard. A single connection point that lets any AI agent , a chatbot, a travel assistant, a booking tool ,tap into your hotel's full ecosystem. Instead of building a separate API integration for each external tool, you connect once to the MCP server, and from there it can reach your booking engine, your PMS, your F&B system, your activities calendar, your spa, your airport transfers. One door in, everything accessible.
The key detail: MCP doesn't store your data. It routes requests to wherever the data actually lives, fetches it in real time, and returns it. No sync layer, no caching issues, no new GDPR exposure. Think of it as a real-time translator with routing rules you define.
What it's not: a magic bridge that makes your fragmented systems suddenly talk to each other. The underlying data still lives in silos. MCP just gives external agents a single, standardized way to query across all of them.
Why This Matters for Direct Bookings
Here's the structural problem MCP addresses. When someone uses an AI assistant to find a hotel today — ChatGPT, Perplexity, Google's AI Overview — that agent defaults to OTA aggregators. Because OTAs have built their APIs. Most independent hotels haven't. So your property shows up stripped of everything that makes it compelling: no mention of your guided fishing trips, your chef's changing menu, your remote location 45 minutes from anywhere, your private dock at sunrise.
MCP changes that equation. A hotel with an MCP server becomes directly queryable by AI agents that support the protocol. Your availability, your rates, your packages, your add-ons — accessible in real time, from your own systems, on your own terms.
The downstream effect: a traveler asking their AI assistant to plan a complete trip accommodation, transfers, activities, meals could receive an itinerary built entirely from your hotel's own data, not from a homogenized OTA listing. And take the booking directly, without a commission going anywhere.
What It Looks Like at a Property Level
Consider a remote lodge running multiple revenue streams: rooms, guided fishing, helicopter rides, transfers from the nearest airstrip, and a dining room with a weekly changing menu. Today, each of those lives in a separate system , a booking engine, a WhatsApp thread, a shared spreadsheet, a handwritten board by the kitchen. A direct booking inquiry still requires a human manually checking all of it.
With MCP, an AI agent could query all of those simultaneously in a single conversation. A guest asks: "I want 4 nights for two people in October, including two guided fishing days, airport pickup from Chaitén, and dinner on arrival." The agent checks availability, rates, and activity slots across every relevant system and returns a complete, bookable itinerary , all from the hotel's own data, not from a third party that flattens everything into a rate comparison.
That's not a distant concept. The technology is functional. The gap right now is adoption: enough properties and PMS vendors need to expose their data through MCP for the ecosystem to reach the point where AI agents actively query it.
The Open Questions
Two structural issues don't have clean answers yet — and they'll determine whether MCP actually delivers.
Where does MCP live? The natural home is your PMS, since it's the most accurate source of truth for availability and pricing. But if PMS vendors own the MCP layer, they also control access and some already charge significant API fees to restrict third-party connections. A neutral, third-party MCP aggregator is the alternative, but that adds a vendor, a cost, and a new dependency. Neither option is clean, and neither has emerged as dominant yet.
Is there a standard? Not really. The protocol gaining the most traction right now is Anthropic's MCP spec, but "gaining traction" in fragmented hospitality tech doesn't mean much until your booking engine actually supports it. Early adopters carry real uncertainty.
The practical move: understand the protocol, track which vendors are moving first, and position yourself to adopt early without building anything from scratch before the standard stabilizes. One question worth asking your PMS provider today: "Do you support MCP, or is it on your roadmap?"
The Longer View
MCP won't rewrite your distribution strategy next quarter. But the direction of travel is clear: AI agents are increasingly how people discover, compare, and book. Hotels that can expose their full ecosystem directly, accurately, in real time will have a structural advantage over those that can't. Right now, that advantage belongs to OTAs by default, because they built the infrastructure first.
The question isn't whether to care about this. It's whether you get ahead of it or react to it once the terms have already been set.
Already experimenting with AI tools in your hotel operations? Tell us what's working we're building a post around real-world hotel AI use cases.