Open-Meteo Weather Scraper — Forecast, History, Air Quality
Pricing
from $0.35 / 1,000 result items
Weather data for AI agents without an API key: current, hourly/16-day daily forecast, ERA5 history since 1940, air quality (AQI, PM2.5, pollen), marine waves, CMIP6 climate projections, ensemble spread, river discharge — for any city or coordinates. 200+ variables, model selection, units, timezone.
Place names (optionally City, CC or City, Country to disambiguate) or lat,lon pairs. Coordinates need one Open-Meteo call per location, a place name one more for the geocoder (remembered for 30 days in your account's open-meteo-weather-geocache store). Names in another script (Cyrillic, Greek, Arabic…) are also looked up in that script's languages.
Which Open-Meteo endpoint and granularity. Each mode has its own variable dictionary (see README); defaults are sensible for the mode.
Forecast days ahead (daily/hourly/marine ≤16, airQuality ≤7, flood ≤210) or days back for history/historyHourly. Ignored when startDate/endDate are set.
Preset; individual unit fields below override it.
Open-Meteo variable names valid for the chosen mode, e.g. daily: temperature_2m_max, precipitation_sum, et0_fao_evapotranspiration; hourly: temperature_2m, soil_moisture_0_to_1cm, shortwave_radiation. Case, spaces, '_' and '-' do not matter. A name the mode does not have is skipped and named in the status and SUMMARY.warnings (with the modes that have it, or a "did you mean"); when no name is valid the run stops before any request. Empty = mode defaults. Full lists in README → Reference.
Forecast modes: best_match (default), ecmwf_ifs025, gfs_seamless, icon_seamless, gem_seamless, meteofrance_seamless, jma_seamless, ukmo_seamless… Several models → one column per model (temperature_2m_max_gfs_seamless). Climate: EC_Earth3P_HR (default), MRI_AGCM3_2_S, CMCC_CM2_VHR4… Ensemble: icon_seamless (default), gfs_seamless, ecmwf_ifs025… Unknown ids are skipped and named in the status; when none is valid the run stops (a typo never turns into a different model).
Ensemble mode: add <var>_values arrays with every member besides min/max/mean/spread.
YYYY-MM-DD. History: archive from 1940-01-01. Forecast modes: up to ~92 days back and 16 ahead. Climate: 1950–2050.
YYYY-MM-DD (inclusive). History defaults to 2 days ago (ERA5 latency).
Prepend N days of recent observations/analysis to a forecast (daily, hourly, airQuality, marine, ensemble, flood).
auto = local time of each location; or an IANA name like Asia/Almaty, UTC.
Overrides the preset.
Language the geocoder matches place names in, and of the returned location label, e.g. ru, kk, de. A name in a non-Latin script that is not found in this language is also tried in its script's languages (Cyrillic: ru, then uk) — the status says which one matched.
ISO-3166 alpha-2, e.g. KZ — restrict all place-name lookups to one country.
Which model grid cell to use for coastal locations.
Air-quality modes only.
Compact is cheaper for long hourly ranges (1 billed item per location). One compact row holds up to ~600 000 values (an item must stay under 9 MB): a longer range is cut and the cut is named in the status.
Hard cap on dataset rows (each row is billed). For date ranges the actor requests only the days that still fit and names the cut in the status and SUMMARY.warnings; locations after the cap are listed as not fetched.
Keep only these columns, in this order (location and date/time are always kept and come first unless you list them elsewhere). Empty = every column. Columns are the base fields in README → Output plus every requested variable under its Open-Meteo name (several models: <variable>_<model>; ensemble: _min, _max, _mean, _spread, _members); with layout compact the variables are inside series. Case, spaces, '_' and '-' do not matter. A name that is not a column is skipped and named in the status and SUMMARY.warnings (with a "did you mean" for near misses); when none of the names is a column, the run stops before any request with the list of columns.
