Station discovery is what makes this more than a forecast lookup: fuzzy matching and geocoding mean an agent can answer a question about a place without first knowing that place's station code, and the climate series reach back decades through the same interface. The output format was measured rather than guessed — across 13 models reading the forecast JSON, hour-level lookups scored ~100% on local-time timestamps against ~0% on UTC, and local time is what ships.
An MCP server over MeteoSwiss Open Government Data, the same data behind the MeteoSwiss app and website. It exposes 7 structured tools with fuzzy station matching, geocoding, and TTL-tiered caching, and there is a hosted instance you can point a client at with nothing installed.
- meteoswissLocalForecast — multi-day forecasts by postal code, station, or place name, across about 6000 Swiss locations
- meteoswissCurrentWeather — temperature, wind, humidity, and pressure from about 300 stations, updated every 10 minutes
- meteoswissStations — find stations by name, canton, or GPS coordinates
- meteoswissPollenData — pollen concentrations from about 15 monitoring stations
- meteoswissClimateData — National Basic Climatic Network series for temperature, precipitation, sunshine, and climate indicators going back decades
- search and fetch — MeteoSwiss website content in German, French, Italian, and English
- Fuzzy matching on station names, so a place spelled the way a person would spell it still resolves
No account and no API key — the data is free. The shortest path is the hosted instance: claude mcp add meteoswiss https://meteoswiss-mcp.ars.is/mcp, or add it from the Cursor Directory. To run it yourself, the npm package is meteoswiss-mcp, or pull the container ghcr.io/eins78/meteoswiss-mcp:latest and publish port 3000. The same repository also ships the identical data access as an agent skill built on curl, awk, and jq, for clients with shell access and no MCP support.
One command — claude mcp add meteoswiss https://meteoswiss-mcp.ars.is/mcp