An open-source project called Travel Hacking Toolkit just dropped. It does something useful. The toolkit lets AI agents like Claude Code and OpenCode search award flight availability across 25+ loyalty programs, compare against cash prices, and tell you whether to burn points or pay. Five free MCP servers work immediately with zero setup: Skiplagged, Kiwi.com, Trivago, Ferryhopper, and a patched Airbnb server. Premium integrations add Seats.aero for award searches, SerpAPI for cash pricing, Duffel for real-time booking, and AwardWallet for loyalty balances. The architecture separates 'skills' (markdown files teaching the AI how to call travel APIs) from MCP servers (actual tools it invokes in real time). This works. Clone the repo, run setup.sh, and your AI agent can answer questions like 'find me a 60,000-mile business class seat to Tokyo in August' by pulling live data from multiple services. You stop tabbing through websites. Full functionality costs money. Seats.aero runs about $8/month for the Pro tier you'd need. SerpAPI offers a free tier with 100 searches monthly. The five free servers need nothing. The real story is the pattern here. MCP infrastructure is solving a genuine multi-API orchestration problem. The "points or cash" question requires checking award availability, cash prices, portal valuations, and loyalty balances across fragmented services. That's the kind of workflow agents should handle.