API for AI Agents — Not Humans

Built for agents first.
Everything else second.

Skyscanner, Amadeus, and Google Flights API were designed for human-facing UIs. AgentTravel was designed from scratch for how AI agents actually call tools. Here's exactly what that means.

Dimension
AgentTravel
Skyscanner API
Amadeus API
Google Flights API
Agent Capabilities
MCP server built-in ✅ ❌ ❌ ❌
Structured JSON responses ✅Normalised, consistent schema ⚠️Inconsistent across endpoints ⚠️SOAP/REST mix, complex nested ⚠️Google-format, opaque fields
Tool-call ergonomics (SDKs) ✅npm i @agenttravel/mcp — done ❌Manual REST + auth boilerplate ❌OAuth 2.0 + token mgmt required ❌Proprietary format, no SDK
Error handling (structured) ✅Machine-readable codes + messages ❌Human-readable HTML errors ❌WSDL errors, cryptic codes ❌Gaxios errors, retry logic missing
Streaming / async responses ✅SSE + polling for long searches ❌ ❌ ❌
Technical Design
Authentication for agents ✅API key + x402 payment token ❌Partner keys, manual onboarding ❌OAuth 2.0, not agent-friendly ❌Google Cloud keys, GCP project
Self-serve sign-up ✅Instant key, no sales call ❌Partnership application required ❌Self-service but opaque docs ❌GCP billing account required
Response normalisation ✅Unified schema across all carriers ❌Raw carrier responses exposed ❌Provider-specific formats ❌Google internal format
Rate limits for agent use ✅Agent-friendly, burst allowed ❌Per-second throttling, no burst ❌Strict hourly caps ❌Heavy on API quotas
Product Coverage
Flight search + booking ✅ ❌Search only, no booking ✅ ❌Search only, no PNR
Hotel + activities ✅ ❌None ⚠️Partial, requires multiple APIs ❌
Ground transport ✅ ❌ ❌ ❌
Points / miles optimisation ✅ ❌ ❌ ❌
AgentTravel
MCP server built-in ✅
Structured JSON responses ✅
Tool-call ergonomics ✅
Error handling ✅
Streaming / async ✅
Auth for agents ✅
Self-serve sign-up ✅
Response normalisation ✅
Rate limits ✅
Flight search + booking ✅
Hotel + activities ✅
Ground transport ✅
Points / miles ✅
Skyscanner API
MCP server built-in ❌
Structured JSON responses ⚠️
Tool-call ergonomics ❌
Error handling ❌
Streaming / async ❌
Auth for agents ❌
Self-serve sign-up ❌
Response normalisation ❌
Rate limits ❌
Flight search + booking ✅ Search only
Hotel + activities ❌
Ground transport ❌
Points / miles ❌
Amadeus API
MCP server built-in ❌
Structured JSON responses ⚠️
Tool-call ergonomics ❌
Error handling ❌
Streaming / async ❌
Auth for agents ❌
Self-serve sign-up ✅ Opaque docs
Response normalisation ❌
Rate limits ❌
Flight search + booking ✅
Hotel + activities ⚠️
Ground transport ❌
Points / miles ❌
Google Flights API
MCP server built-in ❌
Structured JSON responses ⚠️
Tool-call ergonomics ❌
Error handling ❌
Streaming / async ❌
Auth for agents ❌
Self-serve sign-up ❌
Response normalisation ❌
Rate limits ❌
Flight search + booking ✅ Search only
Hotel + activities ❌
Ground transport ❌
Points / miles ❌
Why this gap exists

These APIs were built for human workflows.
Agents don't have browsers.

Skyscanner, Amadeus, and Google Flights optimise for humans clicking through a UI — session tokens, redirect flows, HTML error pages, OAuth popups. Every single one of those assumptions breaks when your agent calls the same API programmatically. AgentTravel was designed for the agent-first case from day one.

Stop working around APIs built for humans.

AgentTravel gives your AI agent the travel stack it actually needs — MCP tools, structured JSON, machine-readable errors, and instant self-serve access.