NWS METAR Weather Observations – Typed Fields, No Parsing

Turn any US airport or weather station into clean, typed weather data: one row per METAR observation, ready for a dashboard, a database, or an AI agent's context window. Give it station IDs (or a US state) and get back temperature, dewpoint, wind, pressure, visibility and humidity as plain numbers — pulled from the official National Weather Service API, free and key-free, with no login and no browser.

Built for data pipelines and dashboards that need reliable current-conditions feeds without babysitting a parser, researchers and analysts who need consistent numeric weather fields across many stations, and AI agents that need to hand a station ID to a tool and get back typed numbers instead of raw METAR text to decode themselves.

Most METAR/NWS scrapers on Apify parse the raw METAR text string (rawMessage), which the NWS API frequently returns empty — that's a likely reason the category's current leader only succeeds on 16% of its runs. This Actor reads the API's own typed JSON fields instead (temperature, windSpeed, barometricPressure, ...), which NWS populates even when rawMessage is blank.

What you get

• One row per observation, with typed numeric columns — temperature_c, dewpoint_c, wind_speed_kmh, wind_gust_kmh, barometric_pressure_pa, sea_level_pressure_pa, visibility_m, relative_humidity_pct, precipitation_last_hour_mm, precipitation_last_3h_mm, precipitation_last_6h_mm — already converted from the API's unit-coded objects into plain floats, null when NWS itself reports no value. NWS populates the 3-hour precipitation field most often; the 1-hour and 6-hour fields are frequently null. After the typed columns, every raw NWS property (temperature, windSpeed, cloudLayers, ... as unit-coded objects) is appended unchanged, so nothing the API returned is lost.
• Two modes: latest (one row per station, its most recent observation) or history (up to N past observations per station, optionally bounded by an ISO start/end window).
• Errors instead of crashes — an unknown station ID becomes a single unbilled error row (error: "not_found", errorMessage), and the rest of your station list keeps going.

Sample row

A real row from a default run against JFK airport:

{
  "station_id": "KJFK",
  "station_name": "New York, Kennedy International Airport",
  "latitude": 40.65,
  "longitude": -73.78,
  "timestamp": "2026-09-11T09:25:00+00:00",
  "text_description": "Mostly Clear",
  "temperature_c": 23.0,
  "dewpoint_c": 13.0,
  "relative_humidity_pct": 53.320072779088,
  "wind_direction_deg": 350.0,
  "wind_speed_kmh": 16.668,
  "wind_gust_kmh": null,
  "barometric_pressure_pa": 101354.61,
  "sea_level_pressure_pa": null,
  "visibility_m": 16093.44,
  "precipitation_last_hour_mm": null,
  "precipitation_last_3h_mm": null,
  "precipitation_last_6h_mm": null,
  "raw_metar": null,
  "url": "https://api.weather.gov/stations/KJFK/observations/2026-09-11T09:25:00+00:00",
  "scraped_at": "2026-09-11T09:49:51+00:00"
}

Note raw_metar is null here — that field only comes back non-empty for stations NWS attaches raw METAR text to, and it's null on plenty of real stations (including this one). Every other field above comes straight from NWS's own typed data, independent of whether rawMessage is populated. The sample is trimmed to the typed columns; a real row continues with the raw NWS properties (see above).

Filters

Input	Default	What it does
stationIds	8 major US airports (JFK, LAX, ORD, DFW, SEA, ATL, DEN, MIA)	Your own NWS/ICAO station identifiers. Ignored when state is set.
state	unset	Optional 2-letter USPS state/territory code (e.g. NY, HI). When set, station IDs are resolved from NWS's own /stations?state=XX list instead of stationIds, capped by maxStations.
maxStations	50	Caps how many stations are resolved when state is set.
mode	latest	latest: one row per station. history: up to maxObservationsPerStation past observations per station.
maxObservationsPerStation	24	History mode only: stop after this many past observations per station, newest first.
start / end	unset	History mode only: optional ISO 8601 bounds, e.g. 2026-09-01T00:00:00Z.

US stations only

This is a direct read of the US National Weather Service API, so coverage is US airports and weather stations only (including territories NWS covers, e.g. Puerto Rico, Guam). It is not a global METAR source.

No personal data

Every field is a government-published weather measurement tied to a fixed physical station, never a person.

Pricing

Pay per event, no subscription: $0.50 per 1,000 observations. A default run (8 stations, latest mode) costs less than half a cent. Error rows (an unknown station ID) are never charged. If every requested station fails, the run is marked failed rather than quietly returning nothing, and you are billed for nothing.

Use it from your tools

• API and SDKs — call it via the Apify API or the official Python/JavaScript clients: one call to start the run, one to fetch the observation rows as JSON, CSV or Excel.
• n8n, Make and Zapier — trigger runs and feed observations into a workflow through Apify's integration for each.
• AI agents and MCP — eligible for agentic use via Apify's MCP server with pay-per-event pricing, so an agent can hand it a station ID mid-task and pay only for the observation it actually got back.

Details

• Fetching: plain httpx GET over https only (no proxy, no browser), with the descriptive User-Agent NWS's API requires. 429 and 5xx responses are retried with exponential backoff (5 tries); 404 is not retried.
• Units: NWS reports every field in a fixed metric unit (°C, km/h, Pa, m, percent, mm); this Actor pulls the number out of the API's {unitCode, value} object and puts the unit in the column name, so no conversion math is needed downstream.
• Run stats: the STATS record in the run's key-value store holds per-run counts (stations, observations fetched/charged, requests, error categories).
• Not affiliated with NOAA or the National Weather Service — this Actor reads their free public API; it is not an official NOAA/NWS product.