What gondola MCP server does
The gondola MCP server connects an AI client to Gondola’s travel search and booking services. It covers hotels, flights, live events, and rental vehicles. Its hotel tools are designed for points-and-miles decisions: an agent can find cash and award rates together, compare their relative value, inspect individual properties, and retrieve a direct booking link.
Rental-car tools support vehicle search, offer details, booking links, and rental insurance checks based on a credit card or a specific booking. Flight search returns cash-priced options with booking links. Event search returns venue, date, and listed-price information.
The service also exposes analytical tools such as price prediction, rate diagnosis, multi-night breakdowns, hotel statistics, comparable-property searches, reviews, and suggested searches. Personal travel and loyalty features in the README include account balances, free-night credits, saved profiles, trips, and rate alerts, although these require sign-in.
How it works
The gondola MCP server is hosted at https://mcp.gondola.ai/mcp and uses Streamable HTTP. An MCP-compatible client sends tool calls to this endpoint and receives search results or booking-related data. Search and discovery calls are available anonymously, so an API key is not needed for basic research.
When a client calls a personal tool, Gondola uses OAuth 2.1 with Dynamic Client Registration. The client opens Gondola’s sign-in flow rather than requiring a manually copied key. This separates public travel searches from account-specific information such as loyalty balances, saved traveler details, and trips.
Hotel and car booking-link tools hand the user a direct Gondola checkout link. Separate partner-only tools can place charges against a saved payment method, but their required mcp:book scope is not granted to ordinary dynamically registered clients.
Setup and configuration
Add the remote URL to an MCP client that supports Streamable HTTP. The README provides configurations for Claude Code, Codex CLI, Cursor, Windsurf, VS Code, Gemini CLI, and Claude Desktop. ChatGPT is also documented through its custom connector flow. The configuration field differs by client: for example, Windsurf uses serverUrl, Gemini CLI uses httpUrl, and VS Code uses an HTTP server entry.
Claude Code can install the repository plugin, which also adds slash commands for hotel points, award flights, redemption evaluation, and points optimization. Direct MCP configuration is available when only the server connection is needed. Anonymous searches can be used immediately; OAuth begins when a personal tool is first invoked.
Tools and capabilities
The gondola MCP server includes tools for:
- Searching hotels, flights, events, and rental vehicles.
- Comparing hotel cash and award prices with cents-per-point values.
- Reviewing hotel details, guest feedback, statistics, similar properties, predicted prices, and nightly rate changes.
- Retrieving hotel and vehicle booking links.
- Inspecting vehicle offers and checking rental-car insurance coverage.
- Diagnosing rates and generating suggested searches.
- Accessing loyalty balances, free-night credits, trips, alerts, and traveler profiles after OAuth sign-in.
- Booking hotels or vehicles, canceling vehicle bookings, and listing saved payment methods where partner booking access is granted.
Limitations and notes
The README distinguishes ordinary self-serve access from partner booking access. Tools that charge a saved payment method require the mcp:book scope and are not normally listed or callable for a dynamically registered client. The supplied tool list includes these partner operations, but availability depends on authorization.
Flight search is described as cash-priced, while hotel search supports both cash and award rates. The server does not require an API key for search, but account-specific features are not anonymous. Booking links lead to checkout rather than indicating that every booking can be completed directly through an unrestricted MCP tool call.
The repository is licensed under MIT. No environment variables are required by the documented remote setup.