The per-provider tools are what lift this past a single forecast: when it matters, you can pull DWD, GFS, ECMWF and Météo-France separately and compare them, or use ensemble_forecast to see the spread rather than one number. geocoding closes the usual gap — most weather tools want coordinates, and here you can hand it 'Paris' and get the latitude and longitude the forecast tools need in the same session, all without a key.
A server over the Open-Meteo weather APIs: 17 tools for forecasts, historical reanalysis, air quality, marine and flood forecasts, elevation, geocoding, seasonal and climate projections, and forecasts from specific national weather models.
- weather_forecast returns hourly and daily forecasts for coordinates, auto-selecting the best model for the location unless you name one
- weather_archive pulls historical ERA5 reanalysis data from 1940 to the present for a date range
- Specialist forecasts: air_quality (PM2.5, PM10, ozone, pollen, AQI, UV), marine_weather (wave height, period, direction, sea surface temperature), and flood_forecast (river discharge from GloFAS)
- geocoding converts a place name or postal code into the latitude and longitude the other tools need, and elevation returns elevation for coordinates, including batch lookups when you pass arrays
- Per-provider models when you want a specific one: dwd_icon_forecast (German DWD), gfs_forecast (US NOAA), meteofrance_forecast, ecmwf_forecast, jma_forecast (Japan), metno_forecast (Norway) and gem_forecast (Canada) — one model per call, so multi-model comparison is parallel calls
- Long-range and uncertainty: seasonal_forecast up to about 7 months out, climate_projection from CMIP6 warming scenarios, and ensemble_forecast showing forecast spread across model runs
Nothing to sign up for — Open-Meteo is free and needs no API key. Node, run as npx open-meteo-mcp-server, or from the ghcr.io Docker image.
One command — npx -y -p open-meteo-mcp-server open-meteo-mcp-server