Grevon launches AI booking automation that skips staff confirmation. A five-unit rental already beat Marriott and Hilton to doing it for real.

TLDR

Marriott's most advanced public AI booking test on record is a chatbot prototype. It was shown to 75 loyalty members in September 2025, not tested as a live transaction, according to Skift.

The industry's first documented hotel booking completed through Model Context Protocol belongs to Rose St. Gardens, a five-unit rental in Dominica. That came months before any branded chain, according to a Hospitality Net opinion piece by GuestCentric CTO Sergio Serra.

Judge "AI booking" vendors like Grevon on whether they can show a completed reservation end to end. Don't judge them on how well they describe their own connectivity.

Grevon, founded by veterans of RMS Cloud, a global hotel property management platform, debuted its AI platform at HITEC 2026 on June 15 in San Antonio. The launch centers on three products built on Kore, an MCP-based connectivity layer. Pulse is a booking agent for hotel websites. Echo is a voice agent for phone reservations. Trade coverage of the launch describes both as completing reservations directly inside a hotel's PMS, not merely answering questions about availability. That claim is worth testing against what has actually been proven in the market this year.

Grevon says its AI agents complete bookings without a staff member confirming them

Grevon launched Pulse and Echo atop Kore at HITEC 2026 on June 15, according to coverage in Hospitality Net, LodgingMagazine and Hotel Business. Pulse "completes reservations directly within a hotel's PMS," per LodgingMagazine's account. Echo "converts calls that would otherwise go unanswered into completed bookings," according to Hotel Business. Neither outlet's coverage says what happens when the AI hits an edge case. Think a rate exception, or a guest who needs a payment plan. That gap matters, because "completes the booking" and "completes the booking every time" are different claims. Only one of them has been demonstrated in public so far.

Picture a 20-room inn that misses eight phone inquiries a week after 9pm, the kind of gap Echo is built to close. Say half of those calls would have converted at a $180 average nightly rate for two nights. That is roughly $1,440 a week going to voicemail or a competitor. An agent that reliably closes even three of those eight calls recovers over $500 a week, with no added staff hours. The upside is real, if the automation holds up unattended overnight. Missed calls cost you nothing you can see. A wrong booking confirmed at 2am costs you a guest.

The first real MCP hotel booking came from a five-unit rental, not a hotel chain

Sergio Serra, CTO of GuestCentric Systems, covered this in a March 13, 2026 Hospitality Net opinion piece. He wrote that the first hotel booking completed through Model Context Protocol did not come from a major chain. "It wasn't a Marriott, it wasn't a Hilton," Serra wrote. "It was Rose St. Gardens, a five-unit short-term rental on the island of Dominica." The booking went through ChatGPT directly, with no OTA in the middle. That timeline puts a five-unit independent property ahead of every branded chain, on the exact capability Grevon is now marketing at scale.

A five-unit property has one owner-operator, deciding alone to plug into a new AI channel. There is no brand IT department or franchise agreement to clear first. A 500-room flagship has legal review, brand standards and loyalty-program integration to work through. All of that has to happen before any AI agent touches a live reservation. Speed to adopt has flipped: the smallest properties are moving faster than the biggest ones. Scale used to be the advantage. On agentic AI, it is turning into drag.

Marriott's most advanced public AI booking test so far did not book anything

Marriott previewed an AI chatbot prototype for its mobile app at a members' event in Bethesda on September 23, 2025, according to Skift. Roughly 75 loyalty members tested the concept. It was built to personalize hotel search, not to complete a transaction. Peggy Roe, Marriott's chief customer officer, led the feedback sessions. The event predates Grevon's HITEC launch by nine months. Marriott still has not made a public claim that its AI can finish a booking unassisted.

A hotel chain testing with 75 loyalty members in a conference room is gathering feedback. A five-unit rental connecting to ChatGPT is running a live transaction, with real money changing hands. One of those is a focus group. The other is a booking. Independent operators sizing up their own AI options are closer to the second category than the brand names suggest.

MCP is an open standard, so Grevon is not the only company building on it

Anthropic published Model Context Protocol as an open standard in November 2024. Coverage of Grevon's own launch notes it is now backed by OpenAI, Google and Microsoft too. Two other MCP-based hotel booking tools reached the market before Grevon's HITEC debut. The Hotels Network, powered by Lighthouse, launched a ChatGPT booking app on March 4, 2026. It routes guests to "the hotel's own website to complete the reservation," stopping short of an autonomous booking. Agentic Hospitality's TravelOS followed on April 28, 2026. Its founder, Brad Brewer, told Hospitality Net, "We are the only company capturing the full conversation and delivering it into the PMS." TravelOS also keeps the booking hotel-controlled, rather than fully autonomous.

If Kore's connectivity layer is genuinely reusable, a hotel could swap Echo for a competitor's voice agent next year. It would not need to rebuild its PMS integration to do it. No vendor owns the pipe, so no single vendor can hold a hotel's data hostage. That is the upside of an open standard. It also means Grevon's real competition is not the protocol. It is whoever executes the booking step most reliably.

Two different things get lumped into "AI booking platform": connectivity and completion

Connectivity is the plumbing. It is an AI tool's ability to pull live rates, availability and guest data from a property's PMS. It does that without a developer writing custom code for each system. MCP standardizes that plumbing, which is why chat and voice vendors keep adopting it. Completion is a separate, harder claim. It means the AI carries out an actual transaction: creating, modifying or cancelling a reservation, with no person checking the result first. A tool can have deep connectivity and still stop short of completion. That is what TravelOS and The Hotels Network's ChatGPT app do, by routing the final step to a human or the hotel's own site.

Hotel tech has already solved this handoff for a narrower job: pricing. PriceLabs, a pricing platform used by independent and boutique properties, connects to more than 160 PMS and OTA systems. It pushes rate changes back into a property's PMS automatically, according to the company's product documentation updated in June 2026. An optional real-time sync can trigger up to 24 automated price updates a day. Those updates tie directly to new reservations or cancellations, with no developer or manual re-entry involved. A team member at Hotel Constantin, a 33-room independent hotel in Germany, described connecting the platform to their PMS as "unbelievably quick and easy." That is the same connectivity-to-completion loop Grevon is now proposing for entire reservations, just narrower in scope. One number updates automatically, instead of a full booking closing with no one checking it first.

Independent hotels are outrunning branded chains on agentic AI, for now

The pattern holds across every MCP hotel booking made public so far. A five-unit rental completed the first one. Boutique and independent properties are the named early adopters in coverage of Grevon's own launch. The specific major chains tied to a live agentic booking are, notably, none. Trade coverage of the HITEC 2026 launch describes early deployments spanning "independent boutiques to multinational hotel groups." But the only chain-level AI news in the same window is Marriott's loyalty chatbot prototype, still in the feedback stage.

That gap has a simple cause. An independent owner-operator can approve a new AI vendor and connect it to their PMS in an afternoon. A branded chain has to clear legal review, franchise agreements and loyalty-program integration first. Any AI agent touching a live reservation has to wait on that process, which runs on a different clock. European independents carry one more layer their US counterparts do not. A chatbot or voice agent handling payment data, or storing guest information, has to satisfy GDPR before it goes live. That can slow a rollout Grevon and its competitors are pitching as a two-minute setup.

For now, the speed advantage sits with the operator who can say yes without a committee. It does not sit with the one holding the biggest loyalty program.

Three things to do this quarter: demand a live demo, check the workflow, and watch the open standard

Ask any "AI booking" vendor to complete a reservation live, not describe one: book a real test reservation on a call or through chat. Watch whether a human touches it before confirmation. If the demo stops at "here's your availability," you are being sold connectivity, not completion.

Ask what happens on the edge case, not the easy case: a rate override, a group booking, a refund request. Grevon's own launch coverage does not specify how Pulse or Echo handle exceptions. That is the exact scenario most likely to need a human anyway.

Track MCP adoption instead of any single vendor's pitch: Model Context Protocol is an open standard, so you are not locked into Grevon, TravelOS or The Hotels Network specifically. Ask your PMS provider whether it supports MCP now, so you can compare tools on execution, not on who reached you first.