If AI booking is going mainstream, why would a small independent hotel let a tech vendor handle it instead of plugging in directly?

This explores why an independent hotel would likely depend on a technology vendor to reach AI assistants instead of setting up its own MCP endpoint. The corpus doesn't study independent hotels directly, so this answer is pieced together from neighbouring evidence. The urgency comes from Will hotels need MCP endpoints to reach AI travel assistants?, which argues that MCP is quickly becoming a booking channel. Bookings are already going through Perplexity, Gemini and OpenAI. But the companies Ly names as moving first are RateGain and Sabre, which are infrastructure vendors, not individual properties. That detail hints at the answer: the channel is opening at the vendor layer.

The engineering notes show why. Exposing an MCP endpoint sounds simple, but production teams report that MCP-mediated tool access caused unpredictable failures. Agents picked the wrong tool or guessed parameters, and reliability only came back after switching to explicit, single-purpose function calls Why do protocol-based tool integrations fail in production workflows?. MCP also leaves conversation state and task lifecycle to the application Does MCP handle multi-turn agent coordination without application code?. A booking spans several turns: search, hold, confirm, perhaps modify. Someone has to write the code that tracks those steps. A 40-room hotel has no engineering team to do that well. A vendor serving thousands of properties does.

The second reason is that an MCP endpoint is only one layer of a larger stack. Agentic commerce needs separate protocols for discovery, governance, payment authorization and agent identity. Merchants stack these layers rather than picking one, and each layer handles a different kind of fraud risk Do agentic commerce protocols compete or complement each other?. Storing credentials securely does not solve the separate problem of agents identifying themselves to merchants Can credential vaulting alone solve merchant-facing agent identification?. Keeping up with all of that is ongoing work, not a one-off project. Vendors can spread that cost across many customers. The same logic shows up in self-hosted LLM economics: break-even depends heavily on volume, and small operators take far longer to recover fixed costs When does self hosting an LLM beat commercial API costs?.

What you might not expect is that this dependence has a price. Discussions of AI shopping agents warn that whoever sits between the buyer and the seller shapes the incentives. An agent funded by merchant referral fees can't be fully loyal to both sides Can an AI agent serve both merchant and user interests fairly?. For independent hotels, that raises a familiar question in a new form. They spent years trying to escape online travel agency commissions. If they now reach AI assistants mainly through vendors, a new intermediary layer could form, and Ly's warning about new intermediaries could come true. The corpus doesn't yet have data on how vendor-mediated MCP access affects hotels' margins or bargaining power. That is a real gap, not something these notes resolve.