Platform · Agent Gateway

Let AI agents book you directly, on your terms

Agent Gateway gives your business its own Model Context Protocol (MCP) server: a catalog agents can read, policies they have to respect, and a checkout with spend limits, confirmation steps and human approval wherever you want it.

Origami birds carrying requests queue through a turnstile at a miniature front desk.
Field noteAI agents are a new kind of guest. They need a front desk built for them, with house rules they can read.

What it does

Travelers are starting to tell AI assistants find us a guided canyon hike for four next Saturday and book it, and assistants are starting to do it. Large distributors are opening agent interfaces to their inventory. An independent operator, lodge or destination without an interface of its own will be met by agents through an intermediary, stripped of its packaging, policies and voice, or not met at all.

Agent Gateway is that interface, for anyone with bookable inventory, from tour and activity operators to destination organizations. It uses MCP to expose what you sell, the rules you sell it under, and a checkout designed for a buyer that can’t be trusted to read the fine print on its own. Your reservation system stays the system of record, and Signals can read the questions agents ask as an early read on demand.

What the gateway exposes, and what it enforces

Agent-facing catalog

Products described for machines as well as people: duration, difficulty, inclusions, meeting points, accessibility and live availability, drawn from the same data your staff use.

Policies as data

Cancellation terms, age and fitness requirements, group limits, weather clauses and what an agent may and may not promise, returned with every offer.

Agent-safe checkout

Time-limited holds, explicit traveler confirmation, spend limits and human approval thresholds. No booking without a summary the traveler actually saw.

Attribution

Every request carries the agent, the key and the surface it came from, so you know which assistants send business and what that business converts to.

Rate limits and quotas

Per-agent limits on searches, holds and bookings, with burst controls, so one over-eager agent can't drain your allotment or your API budget.

Audit logs

Every call, offer, hold, confirmation and refusal is logged and exportable. When someone asks why an agent booked what it booked, you have the answer.

How it works

How an agent books a tour

A simplified trace of one booking through the gateway. Tool names come from our reference implementation; your catalog shapes the details.

  1. 01

    Connect

    An MCP-compatible assistant connects with a key you issued, scoped to the tools you've allowed. Unknown agents can browse; only approved ones can hold or book.

  2. 02

    Search

    The agent calls search_availability with dates, party size and needs, and gets live options from your catalog with prices, policies and accessibility details attached.

  3. 03

    Hold

    create_hold reserves the slot for a short window and locks price and terms. Holds expire on their own. Nothing has been sold yet.

  4. 04

    Confirm with the traveler

    The agent must show the traveler the total, the cancellation terms and the inclusions, then return a confirmation token. Above your thresholds, the request waits for someone on your team.

  5. 05

    Book and record

    confirm_booking creates the booking in your reservation system through Booking Flow, takes payment through your payment provider, records attribution and writes the audit log.

Agent-safe checkout

A gatekeeper at the till

An AI agent is a strange new kind of customer: fast, tireless, occasionally confused, and spending someone else's money. The gateway's checkout treats every agent as a polite stranger holding a permission slip. It gets checked, limited and logged.

  • Spend limits per agent, per traveler and per transaction, with daily ceilings you can change at any time.
  • Confirmation steps that require the traveler-facing summary, not just an agent asserting the user agreed.
  • Human approval thresholds for group size, trip value, unusual requests or first-time agents.
  • Prompt-injection defenses. Product content and agent messages are treated as untrusted data, never as instructions to the checkout.
  • A kill switch for one agent or for all agent traffic, effective immediately.
Read: agent-safe payments
A gatekeeper at a desert toll booth inspects a permission slip held by an origami bird.
Field noteAgent-safe checkout reads the permission slip before the gate lifts: limits, confirmations, and a human above a threshold.

Side by side

Reached through intermediaries, or reached directly

Intermediaries will stay part of the mix. The question is whether agents can also reach you on your own terms.

AspectOnly through intermediariesAlso through your Agent Gateway
Your productFlattened into a generic listingYour packaging, inclusions and voice, as you wrote them
Your policiesSummarized, sometimes wronglyReturned with every offer and enforced at checkout
Price and marginShaped by someone else's commission modelYour rates, your promotions, your rules
The customerBelongs to the platformBooked with you, attributed to the agent that sent them
When it goes wrongA ticket in someone else's queueAn audit log, a kill switch and your own team

We wanted assistants to sell our trips the way we would, with the fitness note and the 48-hour cancellation rule intact. The gateway is the first thing that made that enforceable rather than hopeful.

Founder, multi-day hiking operator in the Mountain West

Questions

Frequently asked questions

What is MCP, and why does it matter to a tour operator?

The Model Context Protocol is an open protocol for connecting AI assistants to tools and data. An MCP server lets any compatible assistant discover what you sell and act on it in a structured way, instead of scraping your website and guessing. Our primer is How to make your tours bookable by AI agents.

Which AI assistants can connect?

Any client that speaks MCP and holds a key you've approved. The gateway doesn't depend on a deal with any one assistant, and you decide which agents get which permissions.

Do agents see our costs or margins?

No. Agents see what you choose to publish: sell prices, availability and policies. Contract rates, margins and internal notes never leave your side of the gateway.

What if an agent books the wrong thing?

The confirmation step makes that rare, and your normal cancellation terms apply. Every booking has a full audit trail, so you can see exactly what the agent was shown and what the traveler confirmed.

What does it take to get ready?

Mostly clean data and clear policies. Our Agentic Commerce Readiness service gets catalog, policies and checkout into shape, and AI Search Visibility helps assistants recommend you in the first place.

Find out how agent-ready you are

Start with the checklist, or bring your catalog to a discovery call. We'll show you what an agent sees of your business today, and what it would see through your own gateway.