Challenge
The travel industry is full of best-in-class APIs that never get used in the open. Most AI assistants planning trips today are flying blind guessing flight times, inventing hotel rates and hallucinating itineraries. The real signal live availability, real pricing, supplier rules lives behind walls.
The client needed a single MCP (Model Context Protocol) layer that any agent could plug into to access trusted travel data through a natural-language interface, while keeping each provider's commercial constraints intact.
Result
JMPR is the bridge. A production MCP server that handles search, availability, structured booking flows and the long-running orchestration agents need to actually finish a plan. Claude Sonnet on top, Trigger.dev underneath, and a typed schema for every provider.
Provider integrations: 12+ travel suppliers wired in through a single typed contract.
Real-time pricing: live availability surfaced into the agent context, not cached guesses.
Structured outputs: every plan is a JSON object the agent and a downstream booking system can both read.
Open by design: MCP-compliant, so any Claude/Cursor/Codex/Anthropic-API client can use it.