Price calendar: SFO -> NRT
61 days of data
Cheapest: $384 (2026-02-12)
Most expensive: $870 (2026-04-13)
Average: $632
2026-02:
12: $384 ***
13: $384 ***
...
20: $540 ***
2026-03:
06: $609 *
...
20: $689
find_nearby_airports
Airports near JFK within 150km:
LGA - LaGuardia Airport (New York, US), 17km away
EWR - Newark Liberty International Airport (Newark, US), 33km away
HPN - Westchester County Airport (White Plains, US), 48km away
How it works
There is no official Google Flights API. This server reverse-engineers Google's internal FlightsFrontendService endpoint. It extracts not just flights, but hidden metadata: emissions, price assessments, daily price calendars, aircraft types, and seat details.
Reliability
Caching: 5-minute TTL cache, identical searches don't hit Google twice
Retry with backoff: exponential backoff + jitter on 429/5xx errors, honors Retry-After headers
Circuit breaker: stops requests after 5 consecutive failures, auto-recovers after 30s
Structural validation: clear errors when Google changes their response format
Price history is stored in SQLite at ~/.google-flights-mcp/prices.db. Each call to track_price records the current cheapest price and reports the trend compared to previous observations.
Development
npm install # Install dependencies
npm run build # Build with tsup
npm test # Run tests (85 tests)
npm run lint # ESLint with FP rules (no-let, immutable-data, no-loop-statements)
npm run typecheck # TypeScript strict type checking
npm run check # typecheck + lint + test (the full CI pipeline)
npm run dev # Build in watch mode
Design
Fully functional TypeScript codebase:
No mutation: all types readonly, no let/var/.push()/for loops
Result monad: errors are values (Result<T, E>), not exceptions
Pure/IO separation: pure transforms are separate from network and disk IO
Composition: pipe() + flatMap() for chaining fallible operations
See CLAUDE.md for architecture details and the Google Flights response format reference.
Limitations
No official API: uses an undocumented Google endpoint that could change at any time. Structural validation makes breakages obvious.
Rate limiting: Google may throttle heavy use. Built-in retry and circuit breaker handle transient issues.
Prices are estimates: prices reflect query time; they can change before booking.
Currency: determined by Google based on IP/locale.
