Flight Recovery MCP
When a flight is cancelled, the airline app tells you the flight is gone but not what else breaks. You check your hotel, your calendar and your notes yourself and work out the damage in your head. Flight Recovery MCP is an MCP server with four tools that an AI agent calls in order: 1. assess_disruption: what happened to the flight (status, cause, timing). 2. map_blast_radius: which commitments break (client dinner, hotel late check-in, morning meeting). 3. rank_alternates: scores replacement flights by priority (protect the meeting, earliest arrival, or lowest cost) and returns one recommendation. 4. plan_recovery: specific ordered actions for the recommended flight, such as moving the airport pickup, notifying the client, and asking the hotel to hold the room. The server makes the decision with fixed rules in code: for example, the earliest flight loses if a tight connection could cost you the meeting. The AI client only explains the result, so the answer is consistent each time. All data is synthetic fixtures from mock airline, commitment and transfer sources. Nothing is booked or sent, and every action is a proposal that needs the user's approval. In production, the mock sources would be swapped for a flight-status API, Google Calendar, a hotel partner API, and an airline rebooking API, while the tools and ranking logic stay the same. Stack: NitroStack (TypeScript), deployed on Render. Repo: https://github.com/dhonicore/flight-recovery-mcp
View details