Personal Agent Protocol for Hotels and Vacation Rentals

A guest asks their agent to find a quiet place for a long weekend. The agent can read about your property. The next questions are harder: how does it reach your live inventory, recognize the guest, and know whether it can act for them?

Personal Agent Protocol addresses the connection between that personal agent and your business. The guest brings an agent. Your business declares the ways it can work with you. Both sides have a say in the access it receives.

For a hotel or vacation rental operator, that creates a path to serve the guest through your own account and booking experience, even when their trip starts in someone else’s AI conversation.

Sierra announced Poppy with Meta and other industry participants. The published design partners include Meta and OpenAI on the personal-agent side, United Airlines and Hertz in travel, and Stripe, Visa, Mastercard, and Adyen among technology providers.

These companies are helping shape the draft. The published list reviewed on October 19, 2026 does not name a hotel or vacation rental operator, leaving room for lodging teams to contribute their needs.

Let an agent search public room and home inventory. Require sign-in before it reads a guest’s reservation. Ask for approval before it commits to a new stay or changes booking terms. Each step gets the access it needs.

The protocol’s introduction uses the same pattern for an airline: search first, account access after sign-in, and approval before purchase. For lodging, an operator could also make eligible member rates available after sign-in, with the booking system checking membership and applying the rate rules.

The operator defines the rates, eligibility checks, and permissions for each task.

Discovery starts with a file at /.well-known/poppy.json on the business’s domain. It describes the organization and the interfaces it offers. That gives an agent a place to learn how to connect.

The protocol then connects the agent’s activity across supported APIs, web browsing, and conversations through a session. When account access is needed, supported sign-in flows establish the guest’s authorization. Direct sign-in uses the business’s own sign-in page.

The distinction matters for hospitality: knowing which guest an agent represents is separate from deciding what it may do. An account connection should not be treated as blanket permission to book, change a stay, or spend money.

MCP gives an AI agent a common way to call tools and read data. In hospitality, those tools can answer questions about properties, availability, quotes, and the next step toward a booking.

Poppy describes how a personal agent connects to a business, including identity and access across interfaces. MCP can be one of those interfaces. A website, an OpenAPI-described API, and a company agent can also have roles.

Together, they let an operator connect property tools to the guest’s authorized session. The agent can carry the same task from a tool call to a website or a conversation with the business.

The optional operations extension ties approval to one revision of an action’s terms, tracks that action across channels, and prevents a retry from performing it twice. That is relevant to a stay where dates, price, or cancellation terms can change before confirmation.

Payments remain an open topic on the path to 1.0, including how Poppy should work with existing payment protocols.

Kismet Agent Pay covers the payment part of Kismet Agent Booking, including supported agent payments and checkout links a guest can finish.

Answer engine optimization and generative engine optimization help a business become a useful source in AI answers. Clear explanations, crawlable pages, consistent property facts, and supported claims give a reader something worth using and citing.

At Kismet, we also use agentic engine optimization for the next step: making a property’s information verifiable and its booking path actionable. Poppy is relevant to that work because it addresses the connection and permissions between an agent and a business.

The two jobs reinforce each other. Explain your properties clearly enough for an agent to understand them. Then give it a supported way to check what is current and continue with the guest’s permission. Measure useful outcomes such as a live quote, a guest reaching checkout, and a confirmed booking alongside mentions and referrals.

Start with the parts a guest’s agent will need regardless of which protocol it uses: clear room and home details, current availability, a dated total, booking terms, and a direct path to confirmation. Keep those facts consistent across your website and the tools that answer live questions.

Ask your technology partners which agents and protocol versions they support, how a guest grants and revokes access, and which actions are actually enabled. Test a full journey, including a changed price or a step that needs the guest to take over.

Ask about the optional feedback endpoint, too. It lets agents report tasks the business does not offer through its connected interfaces. A lodging operator could learn that guests want split payments, a pet add-on, or a way to request a minimum-stay exception. Those reports reveal missing capabilities; the operator still decides which requests to support.

Explore Kismet Agent Booking to see the path from a guest’s conversation to a direct reservation.