When an AI assistant tells you check-in is "commonly around 3 p.m.," you know it read the same generic travel guide as everyone else. When it tells you check-in at Hotel Aurora is 3 p.m. and early check-in from 1 p.m. costs 30 euros, that's something else. That's the property management system talking directly to the language model, and the plumbing that makes that possible is starting to show up in hotel back offices.
Model Context Protocol, the open standard Anthropic released in November 2024 to give AI systems a consistent way to connect to data sources and tools, has reached 97 million monthly SDK downloads as of March 2026. The number alone is noise. What's worth examining is where the downloads are landing.
The infrastructure question hotels finally answered
Hospitality has a chronic integration problem. Property management systems, central reservation platforms, CRM tools, and revenue management software all live in separate silos. Before MCP, connecting an AI assistant to those systems meant writing a custom integration for each one, maintaining it across software updates, and rebuilding it whenever a vendor changed an API. This is the reason 45 percent of hotels cite integration challenges as a barrier to AI adoption, and 51 percent say they lack a clear AI strategy, according to a study by hospitality consulting firm h2c.
MCP doesn't solve strategy. What it addresses is the integration layer itself. The protocol uses a client-server architecture with JSON-RPC 2.0 as the message format, giving hotels a standardized way to wire AI assistants into existing systems without replacing those systems. "MCP is not a system of record," Shiji noted in its analysis of the protocol's hospitality applications. "It does not replace PMS, CRS, or CRM platforms." That's the pitch, and it's a deliberate one: the protocol is designed to fit alongside existing infrastructure, not displace it.
Dreamplace Hotels and the 28x number
The proof that matters is the return-on-ad-spend figure. Dreamplace Hotels, a chain operating vacation properties in Spain and Portugal, achieved 10 percent growth in direct bookings and a 28x return on ad spend using a marketing platform built by Cendyn that integrates with hotel systems via MCP. The chain operates across multiple properties and markets direct booking channels against OTA intermediaries where commission fees typically run 15 to 25 percent per transaction.
Direct booking growth at that scale changes the unit economics of digital marketing for hotels. Cendyn, which serves more than 32,000 hotel customers in over 150 countries, positioned the MCP integration as a way to give AI assistants access to live inventory, pricing, and availability rather than synthetic or historically averaged data. The 28x ROAS figure is the number that will get conference panels and vendor pitches built around it. The more durable significance is what it represents: a hotel chain with real properties, real customers, and a real measurable outcome, using AI infrastructure that connects to its actual operating systems.
The broader adoption signal is harder to dismiss as marketing. Shiji, which provides property management and distribution technology to more than 91,000 hotels worldwide, has published guidance on implementing MCP across its hospitality-specific server implementations. That footprint matters. If Shiji's customers adopt the protocol at meaningful scale, MCP becomes the plumbing layer connecting AI assistants to a meaningful share of global hotel inventory.
From open source to open foundation
MCP's governance shifted in December 2025 when Anthropic donated the protocol to the Linux Foundation's Agentic AI Foundation, established on the same date. The foundation's founding projects are MCP, goose (Block's internal agent framework), and AGENTS.md (OpenAI's agent specification). The founding members reads like a who's who of the current AI landscape: AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI all joined as platinum members.
The move to a foundation is not a neutral technical decision. When an individual company controls a protocol's roadmap, adoption decisions carry the weight of that company's commercial interests. Foundation governance distributes that risk, but it introduces new ones: which member's priorities get precedence when commercial interests conflict? The AAIF's founding members include every major hyperscaler and every major model provider. The protocol that hotels are now plugging into is the same protocol that AWS, Google, and Microsoft have endorsed at the foundation level. For a property management vendor evaluating a three-year infrastructure bet, that concentration is worth noting.
The alternative is writing custom integrations anyway. For 78 percent of hotel chains already using some form of AI, according to the h2c study, the question isn't whether to connect AI assistants to hotel systems. It's whether to do it through a standardized protocol with a broad ecosystem or through bespoke vendor-by-vendor work. MCP is currently the only protocol with meaningful traction across both the hospitality vertical and the broader agent infrastructure stack.
What this means for the people building it
The Dreamplace Hotels result is a single data point from a single chain. Hotel marketing technology vendors have a long history of publishing conversion statistics that are favorable to their products. Readers should hold that caveat. The more interesting signal is the infrastructure pattern: hotels are now treating MCP as production plumbing rather than experimental architecture. Shiji's willingness to publish implementation guidance for its 91,000-property customer base suggests the protocol has crossed a threshold where a major PMS vendor considers it stable enough to recommend.
For developers building on the hospitality stack, the implication is straightforward. If your AI product needs to know whether a room is available, what the early checkout policy is, or what the actual negotiated rate is for a corporate account, MCP is increasingly the integration path that hotel IT teams will recognize. The alternative is another custom connector that becomes technical debt the moment the next software update lands.
The 28x ROAS number is the headline. The 91,000 hotels worth of Shiji footprint is the scale. What's happening in between is the plumbing getting real.
