Most search servers stop at the ten blue links. This one carries on into the work that usually follows a lookup: find the flight, price the dates, find the hotel, read what people said about the restaurant. Read it as a search plumbing layer — careful about caching and failure, and bounded by whatever your SearchAPI.io plan allows.
A Python server that puts SearchAPI.io's Google endpoints in front of an assistant. Web, video and AI-overview search sit alongside Maps, Events, Flights and Hotels, so a question and the booking-shaped work that follows it are answered from one vendor. Underneath, calls are pooled and cached with a configurable TTL, retried with exponential backoff, and cut off by a circuit breaker when the upstream starts failing.
- Google web search with the knowledge graph and answer box, filtered by location, language, time period and safe search
- An AI-overview mode that returns a generated summary with its cited sources, structured content blocks and the ordinary results underneath
- Video search filtered by upload time and device
- Maps place search, a place's own detail record with hours, photos and amenities, and its reviews sorted by date or rating
- Event search for concerts, conferences and festivals, narrowed by a date chip such as weekend or month
- Flight search with class, stops and passenger filters, a price calendar for flexible dates, and airport-code lookup
- Destination discovery by interest and travel period, for when the departure airport is fixed and nothing else is
- Hotel search by rating, class, price range, amenities and free cancellation, down to a single property's detail for given dates
- A clock tool that returns today's date and a run of future dates in the format the travel calls expect
- A health check reporting API connectivity, latency, circuit-breaker state, cache statistics and error counts
Python 3.10 or newer and a SearchAPI.io key in the environment; `uvx --from git+https://github.com/RmMargt/searchAPI-mcp.git mcp-server-searchapi` runs it without installing anything, and the client configurations in the repository cover Claude Desktop, VS Code, Zed and Cline. The searches themselves are billed by SearchAPI.io under your plan, and its rate limits are what the retry logic is backing off from. Everything in a query is sent to SearchAPI.io; responses are cached locally, which you can turn off. On Windows, set the client's environment to UTF-8 encoding or non-ASCII results will fail to render.
One command plus a key — uvx --from git+https://github.com/RmMargt/searchAPI-mcp.git mcp-server-searchapi, then supply credentials