Google Flights Policy Recommender (MCP)

A remote MCP server that searches Google Flights, labels every result in-policy or out-of-policy against a travel policy file, and ranks them by your travel preferences — nonstop, airline alliance, red-eye avoidance, cheapest/fastest/balanced. It never visits an airline or OTA site and never attempts to book anything: it hands you a ranked, labeled list and a Google Flights link to complete the booking yourself.

Built as a technology demonstration of policy-aware flight recommendations.

There's also a local, browser-driven server in this repo that can drive a booking up to (but never through) payment — see local-server/README.md.

Connect it to claude.ai

The live deployment:

https://corp-travel-mcp-508362522869.us-central1.run.app/mcp

1. In claude.ai, go to Settings → Connectors → Add custom connector.
2. URL: paste the URL above.
3. Auth: choose OAuth. Leave the Client ID / Client Secret fields blank — this server supports dynamic client registration, so claude.ai registers itself automatically; nothing to fill in by hand.
4. Save. claude.ai will prompt you to connect — sign in with your @routespring.com (or @gallop.ai) Google account. Any other domain is rejected.
5. Try asking it to find a flight (see docs/examples.md).

Running your own separate deployment instead? See docs/advanced-build.md for deploying it (Docker, Fly.io/Render/Cloud Run) and setting up your own Google OAuth client — or use the simpler username/password (Basic Auth) option claude.ai also offers, where the password is the deployment's MCP_BEARER_TOKEN.

Change the default policy & config

Two files drive every recommendation: policy (hard/soft rules — fare caps, cabin class, blocked airlines) and preferences (personal taste — nonstop, alliance, cheapest vs. fastest). Full schema for both: docs/advanced-usage.md.

• Change the deployment's default — edit policy.json / user_preferences.json and redeploy, or set the POLICY_JSON / PREFERENCES_JSON env vars directly (no rebuild needed). This changes results for everyone using that deployment — see docs/advanced-build.md for exact commands.
• Use your own policy without changing the shared default — keep a personal policy.json / user_preferences.json and just mention it ("use my policy for this"); the assistant reads it and passes it in for that call only. This is also how multiple colleagues share one deployment with different policies — see docs/advanced-usage.md.

More

• docs/examples.md — worked examples: a search, checking the active policy, bringing your own policy, booking.
• local-server/README.md — the browser-driven server that can autofill a booking up to payment.
• docs/advanced-build.md — deploying, Google OAuth setup, running it locally.
• docs/advanced-usage.md — full policy/preferences schema, the multi-colleague pattern, known limitations, safety boundary.