Your own MCP interface
An MCP server, delivered through Agent Gateway, that lets assistants search, read policies, hold and book against your live systems.
Services · Agentic Commerce Readiness
Ready for the traveler who sends an agent to book
We make your inventory, policies and checkout readable by AI agents through your own Model Context Protocol (MCP) interface. Assistants can then book you directly, on your terms, with a person in the loop where it counts.
The traveler’s assistant is already shopping
A traveler types: book us a sunrise balloon flight for two next Saturday, with free cancellation if the wind is up. The assistant goes looking and finds a website built for people. Availability sits inside a date picker it can’t operate reliably. The weather policy is paragraph four of a PDF. Checkout wants a card, a CAPTCHA and a human. So the agent books whoever it can read. Usually that’s an intermediary, at their commission, under their brand.
Distribution is being rebuilt for exactly this traveler. In May 2026, Skift reported that Expedia Group was preparing a B2B MCP server so partners’ AI agents could connect directly to its inventory, and lastminute.com has opened an MCP flight server to AI models. If agents can reach you only through someone else’s interface, they’ll describe and sell you on someone else’s terms.
What readiness actually means
Readiness isn’t a chatbot or a plugin. It comes down to four plain properties. An agent can find what you sell, understand the rules, transact within limits you set, and you can see what it did. Each one is engineering work across inventory, policies, checkout and reporting. It matters most for experiences and activities, lodges and outdoor operators with time-slotted, bookable inventory.
What we deliver
Six workstreams, sized to your inventory.
Agent-facing catalog
One clean record per bookable thing, described the way agents compare: duration, difficulty, meeting point, inclusions, accessibility, availability and price.
Policies as data
Cancellation tiers, deposits, weather calls, minimum ages, pet and luggage rules rewritten as structured fields with effective dates, so agents quote them instead of paraphrasing.
Agent-safe checkout
Scoped permissions, spend limits, expiring holds and traveler confirmation.
Attribution
Every agent-originated booking tagged with the assistant, session and product, and reported next to your direct and intermediary channels.
Assistant testing
Scripted booking journeys run through the major general-purpose assistants and agent frameworks, with failures logged, fixed and re-run before launch.
Agent-safe checkout, without handing over the keys
Search is the easy part of agentic commerce. Money is the hard part. We design checkout so an agent does the tedious work and a person still owns the commitment. You set the rules per product and per channel, and the gateway enforces them.
- Holds before payment: the agent reserves the seats, and the hold expires on a clock you set
- Traveler confirmation: the traveler approves the final price and policy before anything is charged
- Scoped, revocable permissions: agents can search widely but book only what you expose, within spend and party-size limits
- Duplicate protection: a retry never creates a second booking or a second charge
- Exceptions to people: group requests, odd dates and unusual policies route to your team through Booking Flow
How it works
From audit to first agent booking
Usually run as an 8–12 week pilot.
- 01
Readiness audit
2 weeksWe ask assistants what they currently know about you and try to book you through them. Then we audit inventory, policies, checkout and the systems behind them against our agent-ready checklist.
- 02
Catalog and policy modeling
2–3 weeksProducts become clean agent-facing records and policies become structured, dated fields. Your team makes the judgment calls.
- 03
Gateway and checkout
2–3 weeksWe connect Agent Gateway to your reservation system or channel manager, set permissions and limits, and route confirmations into the workflow your staff already use.
- 04
Assistant testing
2 weeksDozens of scripted journeys: the easy booking, the family with a toddler, the cancellation, the agent that retries three times. We fix what breaks and re-run.
- 05
Controlled launch
1–2 weeksWe open the interface for a few products, watch attributed bookings and exceptions daily, then widen.
Side by side
What changes for an agent looking at you
AspectBefore readiness workAfter readiness work
How agents find youThrough intermediaries' listings, at their commissionDirectly, through your own MCP interface
AvailabilityScraped, cached or guessedQueried live from your system of record
PoliciesParaphrased from a PDF, sometimes wronglyRead from structured fields with effective dates
CheckoutBlocked for agents, or open with no limitsHolds, limits and traveler confirmation you control
MeasurementAgent bookings invisible in your reportsAttributed by assistant, product and session
Checklist
Outcomes to expect
By the end of the engagement, each of these should be true, and we'll have shown you each one working.
0 of 5 ready
The first test run was humbling. One assistant quoted our old cancellation policy from a cached listing, and another couldn't get past the date picker. A few weeks later both were booking our sunrise flights correctly, and we could see exactly which bookings they brought.
Platform modules and services behind the work
- PlatformAgent GatewayThe MCP server, agent-facing catalog and agent-safe checkout this service configures.
- PlatformBooking FlowConfirmations, vouchers and the exception queue that agent bookings flow into.
- PlatformIntegrationsHow we connect to reservation systems, channel managers and payment providers.
- ServiceAI Search Visibility for TravelThe companion service: getting recommended before getting booked.
Questions
Frequently asked questions
What is the Model Context Protocol, in plain terms?
MCP is an open protocol for connecting AI assistants to tools and data. For a travel business, it means an assistant can ask your system structured questions (what's available, what it costs, what the rules are) and take permitted actions, instead of guessing from web pages. Our MCP primer goes deeper.
Do we have to leave the OTAs and intermediaries we sell through?
No. Readiness adds a direct agent channel beside your existing distribution and removes nothing. Most clients keep intermediaries for reach.
Can an agent book and pay without a person involved?
Technically yes, if you allow it. Most clients start with holds plus traveler confirmation above a set value, and loosen limits only after watching the channel behave. The rules are yours, and the gateway enforces them.
Which assistants do you test with?
The major general-purpose assistants and agent frameworks that support MCP or tool calling at the time of your engagement. That list changes month to month, so ongoing re-testing is part of Managed AI Operations if you want it.
What if our reservation system has no API?
We've worked with systems that only export CSV or confirm by email. Agent Gateway can sit in front of a scheduled feed plus a request-and-confirm booking flow. It's slower than a live API, and the agent is told so honestly.
Related
Keep exploring
- PlatformAgent GatewayYour own MCP server and agent-safe checkout, so AI assistants can find and book you directly under your policies, spend limits, attribution and audit logs.
- ResourcesAgent-ready checklistA 49-item checklist for travel suppliers preparing for AI agents: content, policies, live availability, safe checkout, permissions, monitoring and legal terms.
- Field Notes · Agentic TravelHow to Make Your Tours Bookable by AI Agents: An MCP Primer for Travel SuppliersA practical Model Context Protocol primer for tour and activity suppliers: which tools to expose, which policies to encode, and how to keep agent checkout safe.
- Field Notes · Agentic TravelAgent-Safe Payments: Letting AI Agents Book Without Losing ControlHow travel suppliers can let AI agents hold and pay for bookings safely, using mandates, quote locks, step-up confirmation, idempotency and an audit trail.
- ResourcesDevelopersREST API, webhooks and MCP access to itineraries, routes, quotes, bookings and signals. OAuth 2.0, environment-scoped keys, sandbox access by request.
Find out what an agent sees when it looks at you
On a discovery call we'll run a few live prompts against your business and walk through what the audit would cover.