MeteoSwiss Open Data

Access Swiss weather data from MeteoSwiss Open Government Data. Free, no API key. All data from data.geo.admin.ch. CSVs use semicolon (;) delimiters. Metadata CSVs are Latin1 — pipe through iconv -f latin1 -t utf-8.

Quick Reference

STAC API base: https://data.geo.admin.ch/api/stac/v1

1. Get Current Weather

<!-- mcp-tool: meteoswissCurrentWeather -->

# Get weather for Zurich (station SMA) — key columns: tre200s0 (temp °C),
# ure200s0 (humidity %), rre150z0 (precip mm), fu3010z0 (wind km/h)
curl -s 'https://data.geo.admin.ch/ch.meteoschweiz.messwerte-aktuell/VQHA80.csv' \
  | awk -F';' 'NR==1 || $1=="SMA"'

Full parameter list in ${CLAUDE_SKILL_DIR}/REFERENCE.md. Missing values appear as empty fields or -. Timestamps are YYYYMMDDHHmm in UTC.

2. Find Stations

<!-- mcp-tool: meteoswissStations -->

# Search weather stations by name (Latin1 encoded)
curl -s 'https://data.geo.admin.ch/ch.meteoschweiz.ogd-smn/ogd-smn_meta_stations.csv' \
  | iconv -f latin1 -t utf-8 \
  | awk -F';' 'NR==1 || tolower($0) ~ /zurich/'

Columns: station_abbr, station_name, station_canton, station_height_masl, station_coordinates_wgs84_lat, station_coordinates_wgs84_lon.

# Search forecast locations (~6000 points: stations, postal codes, mountains)
# NOTE: Asset key has a known typo "forcasting" — always get URL from STAC
META_URL=$(curl -s 'https://data.geo.admin.ch/api/stac/v1/collections/ch.meteoschweiz.ogd-local-forecasting' \
  | jq -r '[.assets | to_entries[] | select(.key | contains("meta_point")) | .value.href] | first')
curl -s "$META_URL" | iconv -f latin1 -t utf-8 \
  | awk -F';' 'NR==1 || $3 ~ /8001/'  # search by postal code

Point columns: point_id, point_type_id (1=station, 2=postal_code, 3=mountain), postal_code, station_abbr, point_name.

3. Get Forecasts

<!-- mcp-tool: meteoswissLocalForecast -->

Two steps: get the latest STAC item, then download parameter CSVs.

# Step 1: Get latest forecast item WITH assets (a brand-new item can exist with its
# assets not yet uploaded — skip those or every parameter comes back empty)
ITEM=$(curl -s 'https://data.geo.admin.ch/api/stac/v1/collections/ch.meteoschweiz.ogd-local-forecasting/items?limit=10' \
  | jq -r '[.features[] | select((.assets | length) > 0) | .id] | sort | reverse | .[0]')

# Step 2: Get a parameter CSV (e.g., daily max temperature for Zurich station, point_id=71)
ASSET_URL=$(curl -s "https://data.geo.admin.ch/api/stac/v1/collections/ch.meteoschweiz.ogd-local-forecasting/items/$ITEM" \
  | jq -r '[.assets | to_entries[] | select(.key | contains("tre200dx"))] | sort_by(.key) | last | .value.href')
curl -s "$ASSET_URL" | awk -F';' 'NR==1 || $1=="71"'

Station forecasts (point_type_id=1) have official daily params: tre200dx (max temp), tre200dn (min temp), rka150d0 (precip), jp2000d0 (weather icon). Every point type (stations, postal codes, mountains) also has hourly params: tre200h0 (temp), rre150h0 (precip), sre000h0 (sunshine), fu3010h0 (wind speed), fu3010h1 (wind gust) — plus jww003i0 (weather icon), which is 3-hourly, not hourly, and is used for daily icon selection rather than an hourly reading series. Timestamps are YYYYMMDDHHmm in UTC, so grouping on the first 8 chars aggregates by UTC calendar day. The MCP server instead buckets hourly readings by the local Europe/Zurich calendar day (DST-aware), so a UTC-day grouping and the MCP tool's hourly[] can disagree for late-evening UTC hours that fall on the next local day — convert to Europe/Zurich time first if you need exact parity with the MCP tool's daily buckets. For stations, prefer the official daily params for temp/precip (a separately-curated MeteoSwiss product that may not exactly match summing/averaging the hourly series); use the hourly params for sunshine/wind (no official daily product) and for an hourly breakdown. Common station point_ids: Zurich (SMA)=71, Bern (BER)=78, Geneva (GVE)=58.

Weather icon SVGs: https://www.meteoschweiz.admin.ch/static/resources/weather-symbols/{CODE}.svg — replace {CODE} with the numeric icon code (e.g., 1.svg for sunny, 101.svg for clear night). See ${CLAUDE_SKILL_DIR}/REFERENCE.md for all codes.

4. Get Pollen Data

<!-- mcp-tool: meteoswissPollenData -->

# Stations: PBE, PBS, PBU, PCF, PDS, PGE, PJU, PLO, PLS, PLU, PLZ, PMU, PNE, PPY, PSN, PZH
# Use lowercase in URLs
curl -s 'https://data.geo.admin.ch/ch.meteoschweiz.ogd-pollen/pzh/ogd-pollen_pzh_d_recent.csv' \
  | iconv -f latin1 -t utf-8 \
  | awk -F';' 'NR==1{print} {last=$0} END{print last}'

Columns: station_abbr, reference_timestamp, then pollen parameter codes (e.g. kabetud1=birch, khpoacd1=grasses) in particles/m³. See ${CLAUDE_SKILL_DIR}/REFERENCE.md for all codes.

5. Get Climate Data

<!-- mcp-tool: meteoswissClimateData -->

Homogeneous climate series from the NBCN network (29 climate + 46 precipitation-only stations), going back decades. Monthly (_m) and yearly (_y) files hold all history; daily is _d_recent (last ~2 years, _d_historical for older).

# Yearly climate summary for Zurich (SMA) — key columns: ths200y0 (mean temp °C),
# rhs150y0 (precip mm), shs000y0 (sunshine min), ths30xy0 (heat days)
curl -s 'https://data.geo.admin.ch/ch.meteoschweiz.ogd-nbcn/sma/ogd-nbcn_sma_y.csv' \
  | awk -F';' 'NR==1 || NR>2 {print}' | tail -5

Monthly columns use m in place of y (e.g. ths200m0); daily uses d (ths200d0). Timestamps are DD.MM.YYYY HH:MM. Station list: STAC ch.meteoschweiz.ogd-nbcn → asset ogd-nbcn_meta_stations.csv (Latin1). Precipitation-only stations live in ch.meteoschweiz.ogd-nbcn-precip with the same URL pattern. See ${CLAUDE_SKILL_DIR}/REFERENCE.md for all parameter codes.

Error Handling

• Station not found: Check metadata CSV (Section 2) for valid abbreviations
• Empty data: Station may be offline — try a nearby station
• 403/404 on pollen: Verify abbreviation is lowercase and is a pollen station (16 total; see the list in the pollen section)
• Garbled text: You're reading Latin1 as UTF-8 — add iconv -f latin1 -t utf-8

Bundled Scripts

Token-efficient CLI tools that output structured key=value pairs. Use these instead of raw curl when available — they handle encoding, error checking, and output parsing.

${CLAUDE_SKILL_DIR}/scripts/current-weather.sh SMA          # current weather for Zurich
${CLAUDE_SKILL_DIR}/scripts/search-stations.sh zurich        # find weather stations
${CLAUDE_SKILL_DIR}/scripts/search-forecast-points.sh 8001   # find forecast point_id by postal code
${CLAUDE_SKILL_DIR}/scripts/forecast.sh 71                   # forecast for Zurich (point_id=71)
${CLAUDE_SKILL_DIR}/scripts/pollen.sh PZH                    # pollen data for Zurich

All scripts accept --help for usage details. Requires: curl, awk, iconv. Forecast also needs jq.

MCP Server Alternative

For complex queries (fuzzy search, geocoding, structured JSON), use the MCP server: claude mcp add meteoswiss https://meteoswiss-mcp.ars.is/mcp

Full Reference

See ${CLAUDE_SKILL_DIR}/REFERENCE.md for all parameters, weather icon codes, and STAC collections.