Will hotels need MCP endpoints to reach AI travel assistants?

Henry Ly argues that Model Context Protocol (MCP) — a standard that lets an AI assistant query and transact against a hotel's live inventory directly — is becoming the channel through which travelers book rooms, and that hotels which have not deployed an MCP endpoint will be excluded from it. He writes that "this is not a concept demo": Perplexity already completes end-to-end hotel bookings through Tripadvisor and Selfbook partnerships, Google has committed to transactional trip planning inside Gemini and AI Mode with Booking.com, Expedia, Marriott, and IHG, and OpenAI is opening its app marketplace to travel companies. He cites a 2026 TakeUp survey finding 90% of US travelers are already aware of AI travel assistants, and predicts that "by the end of 2026, MCP-enabled travel bookings will be live across multiple AI platforms, not theoretical, not demo-only."

The mechanism he gives is economic and architectural. OTAs hold roughly 55% of hotel booking market share and charge 15-25% commission per booking — "$30 to $50 going to the intermediary" on a $200 room every night — while, as he puts it, "MCP takes zero." Architecturally, MCP "sits alongside" a hotel's existing PMS and CRS as an orchestration layer rather than replacing them: a hotel wraps its existing systems in standardized tool definitions once, and any MCP-compatible AI assistant can connect to that one integration. He quotes citizenM CIO Mike Rawson describing this as a shift "from coding to describing intent." Because most independent hotels and small chains lack engineering teams and depend on their PMS or booking-engine vendor for distribution technology, Ly frames the real bottleneck as awareness and vendor support, not the technology itself — pointing to RateGain's 2025 MCP-enabled booking engine and a February 2026 Sabre/PayPal/MindTrip partnership as evidence the industry is already moving.

Ly's claim is squarely about MCP as the integration layer connecting an assistant to a merchant's live inventory, which sharpens Does MCP handle multi-turn agent coordination without application code? with a concrete commercial use: the protocol's job here is tool-level read/write access to rates and availability, not multi-turn task state, which the booking flow (PMS, CRS, PayPal) still carries. It also instantiates the general dynamic in Why do consumers delay adopting agentic search? — Ly's version is that merchants who deploy early capture the channel before new intermediaries (or OTAs reasserting themselves through AI partnerships) can claim it, so the adoption lag risk here falls on the merchant rather than the platform. That optimism about MCP as transactional infrastructure sits in tension with Why do protocol-based tool integrations fail in production workflows?, which found MCP integration itself produced ambiguous tool selection and inconsistent parameter inference in production; Ly's essay gives no detail on how booking-critical fields like rate and availability accuracy are guaranteed reliable through an MCP layer at scale.

The excerpt does not establish that MCP adoption among hotels is actually happening at the pace claimed; the cited evidence is a handful of named partnerships and vendor announcements (RateGain, Sabre/PayPal/MindTrip, Skift and OAG reporting) rather than measured uptake, and the "by the end of 2026" prediction is the author's own projection. Nor does it address whether OTAs themselves will simply become the AI assistants' backend data source — as the piece itself notes hotels without MCP will be represented by third-party data... often filtered through an OTA that takes a commission anyway — which would blunt the disintermediation it predicts. The implication, at the strength the evidence allows, is that MCP lowers the technical barrier to disintermediating OTAs but does not by itself guarantee hotels capture that value instead of a new layer of AI-platform intermediaries.