Google Flights

google-flights-mcp

The most feature-rich MCP server for Google Flights. No API key required.

12 tools for searching flights, tracking prices, comparing cabins, analyzing layovers, and more, all powered by real-time Google Flights data.

Works with Claude Desktop, Claude Code, Cursor, and any MCP-compatible client.

Features no other Google Flights MCP has

• Carbon emissions per flight and per leg (CO2 in kg)
• Price context: Google's own assessment of whether a price is low, typical, or high
• Price tracking: track prices over time with SQLite persistence, detect drops and trends
• Calendar heatmap: the cheapest fare for each day of a window of up to 31 days
• Cabin class comparison: economy vs premium vs business vs first, side by side
• Nearby airport suggestions: find cheaper alternatives (e.g., EWR/LGA near JFK)
• Layover analysis: connection time risk assessment (tight/comfortable/overnight)
• Aircraft & seat details: Boeing 787 vs 777, seat pitch in inches
• Google Flights URLs: direct booking links users can click

What you can ask

• *"Find nonstop flights from JFK to London next month"*
• *"What's the cheapest week to fly SFO to Tokyo?"*
• *"Compare economy vs business class for LAX to Paris"*
• *"Are there cheaper airports near JFK I should check?"*
• *"Track the price of SFO to NRT on June 15 and alert me if it drops"*
• *"Show me the emissions for flights from SFO to Tokyo"*
• *"Analyze the layovers for connecting flights from SFO to Bangkok"*

Quick start

Option 1: npx (no install)

Works immediately with Claude Code:

claude mcp add google-flights -- npx -y google-flights-mcp

Or with Claude Desktop, add to your claude_desktop_config.json:

{

"mcpServers": { "google-flights": { "command": "npx", "args": ["-y", "google-flights-mcp"] } } }

Option 2: Install from source

git clone https://github.com/andreacappelletti97/google-flights-mcp.git

cd google-flights-mcp npm install npm run build

Then connect to Claude Code:

claude mcp add google-flights node /absolute/path/to/google-flights-mcp/dist/index.js

Or Claude Desktop (claude_desktop_config.json):

{

"mcpServers": { "google-flights": { "command": "node", "args": ["/absolute/path/to/google-flights-mcp/dist/index.js"] } } }

Restart your MCP client. You should see 12 flight tools available.

Tools (12)

Search

Price intelligence

Airport tools

Utility

Example output

search_flights (with emissions and price context)

Flights from SFO to NRT on 2026-06-15:

Price assessment: HIGH ($294 above typical). Range: $500 - $700, typical: $577

Flight 1: $874 | 11h 15m | 0 stop(s) | CO2: 428kg ZG 25: SFO -> NRT (11h 15m) [aircraft: Boeing 787, seat pitch: 31 inches, CO2: 428kg]

Flight 2: $1310 | 11h 5m | 0 stop(s) | CO2: 545kg NH 7: SFO -> NRT (11h 5m) [aircraft: Boeing 777, seat pitch: 34 inches, CO2: 545kg]

get_calendar_heatmap

Price calendar (one-way, cheapest per day): SFO -> NRT

14 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 loads Google Flights' public search page and reads the result data the page embeds, the same data Google's internal FlightsFrontendService returns. 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
• Paced fan-out: round trips and calendars load several pages; they go out a few at a time (GF_MCP_MULTILEG_CONCURRENCY, GF_MCP_CALENDAR_CONCURRENCY)
• TLS mitigation: undici with browser-like headers + rotating User-Agents, fetch fallback

Price tracking

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 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 ), 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: reads data embedded in Google's public search page, which can change at any time. Structural validation makes breakages obvious.
• Multi-city is unavailable: since August 2026 Google's internal flight endpoints only accept requests signed by its own page JavaScript, and the public page carries no rows for multi-city trips.
• Calendars cost one page per day: Google's single-call calendar endpoint is gated the same way, so get_calendar_heatmap and get_price_insights search each day, up to 31 days.
• Not encoded: maxPrice and maxDuration are applied to the results after the search, so a capped search can return fewer options.
• 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.

Contributing

1. Fork the repo
2. Create a feature branch
3. Make your changes
4. Run npm run check to verify
5. Open a PR

License

ISC