Skip to main content
OnArrival

Travel search and booking tools for AI agents.

Give an MCP-capable agent real travel search, pricing, booking and refund tools. Each key decides what the agent may do, how much it may spend and when a person has to approve.

At a glance

Who it’s for

Teams building AI assistants and agents that need to search and book real travel.

The problem today

An agent that books travel needs real booking tools, spending limits and a record of every step.

With OnArrival

Connect the MCP server and your agent gets travel tools, with limits set per key.

Next step: Scope an agent key

Key facts

  • Allowed tools and spend limits set per key
  • Human approval above a threshold you set
  • Every tool call written to the audit trace
  • SOC 2 Type II, audited annually

Terms in plain English

MCP
Model Context Protocol: the open standard AI agents use to call outside tools.
Tool call
One action the agent asks for, such as a flight search.
Key
The credential that sets what one agent is allowed to do.
Why agents run on us

Typed tools with explicit permissions.

01
Native MCP server
Real tools such as flights.search, policy.check and bookings.create, each with a schema the agent reads and calls directly.
02
Guardrails as code
Limits per traveller and policy are enforced on our servers. The agent can only do what you allow.
03
Audit-grade trace
Every tool call, decision and booking is logged and cannot be edited: you can replay it, export it and see who did what.
How an agent books

Connect the tools. Set booking permissions.

Add the MCP server to your agent, then choose the tools and spending limits it gets.

THE AGENT LOOP · TRACED END TO ENDagentintenttool callflights.searchgatepolicy.checkticketedbookings.createbooking.create · signed · 142msTRACE · immutable · replayabletrace
  1. Connect

    Add the MCP server

    One config block connects OnArrival to Claude Desktop, your IDE or any MCP-capable runtime.

  2. Reason

    Agent searches and prices

    A plain-language request becomes tool calls across flights, stays and more, priced live.

  3. Guard

    Policy checks first

    Budget, cabin and traveller rules are checked before any booking tool can run.

  4. Book

    Confirm and trace

    Supplier confirmations are checked, and every tool call, approval and outcome goes into your audit log.

In the loop

Watch it book, step by step.

A request becomes tool calls for search, policy checks and booking. Each call stays visible and attributable.

Plain language in

“Book the cheapest refundable non-stop, under budget.”

Tool calls out

flights.search → policy.check → bookings.create.

Every step traced

Inputs, outputs and approvals logged for replay.
Limits you set

The agent only does what you allow.

Spending authority lives in rules on our servers, not in the agent. Each key has a list of allowed tools, a spending ceiling, an approval threshold and an off switch.

Scoped keys

Each key lists its allowed tools, from search-only to booking under a cap.

Hard spending limits

Checked before a booking is committed.

Human approval

A lower threshold can pause an action for approval in your app. Approval never overrides the hard limit.

Revocation

Revoke a key and that agent's next tool calls are blocked.
Capabilities

Tools, permissions and audit controls.

Tools

What your agent can do.

  • Search and price
  • Book and cancel
  • Refund and exchange
  • Manage bookings

Runtimes

Where it runs.

  • Claude Desktop
  • GPT tool calling
  • LangGraph and custom agents
  • Local or hosted

Guardrails

Limits you control.

  • Budgets per traveller
  • Policy checked at each tool
  • Spend and rate caps
  • Human approval steps

Trust

A record you can check.

  • Trace log that cannot be edited
  • Replayable sessions
  • Personal data redacted
  • Safe retries
Works with

Bring your own agent.

Connect any MCP-capable runtime. The agent and its conversation stay in your environment.

Claude

Desktop and API, native MCP.

GPT

Responses API and tool calling.

Your agent

Any MCP-capable runtime.
Add to your runtime

Connect the MCP server to your agent.

For Claude Desktop, add the server to claude_desktop_config.json under mcpServers and restart: the travel tools appear. Limits live on the key, not in the config.

claude_desktop_config.json
"mcpServers": {
  "onarrival": {
    "command": "npx",
    "args": ["-y", "@onarrival/mcp"],
    "env": { "ONARRIVAL_KEY": "sk_live_••••••••" }
  }
}
zsh — install
npx -y @onarrival/mcp
First call
  • flights.search → typed offers back
  • policy.check → pass / fail with reason
  • bookings.create → real PNR, idempotent
  • No travel-domain code on your side
Get started
Review MCP permissions and limits.
Related

Pairs well with

Most of our customers use two or three of these together.

Related reading

All field notes →
AI and agents · 12 min

AI travel agents need booking permissions and spending limits

An AI agent can find a trip without having permission to buy it. Booking needs separate checks for the approved price, traveller, budget and any repeated request.

Pawan ShettyTechnical · 11 min

What to monitor after a travel booking is confirmed

Search latency cannot tell you whether a flight changed or a refund arrived. Post-booking monitoring needs to detect supplier changes, missing updates and delays in telling the traveller.

Manan Vaghasiya