MeteoSwiss Open Data → Python API, CLI, MCP server, Parquet & Delta tables

foehn downloads every MeteoSwiss OGD collection via the STAC API, converts CSV/TXT to Parquet with Polars, and optionally ingests everything into Databricks Unity Catalog Delta tables on a daily schedule. It also ships an MCP server so LLMs can query Swiss weather data directly.

Why foehn?

• 20+ collections in one command — weather stations, radar, hail maps, forecasts, climate scenarios, and more
• MCP server for LLMs — give your favorite LLM live access to MeteoSwiss data with the MCP server
• Significantly smaller on disk — columnar Parquet with Zstandard compression vs. raw CSVs
• Incremental by default — only re-downloads files that changed since your last run, tracked via _last_run.json
• No Spark required locally — download + conversion uses Polars only; Spark is optional for Delta ingestion
• Ships a Declarative Automation Bundle — ready-to-deploy daily job and historical backfill, no pipeline config needed

Quick start

pip install foehn
foehn download

Recent data (Jan 1 → yesterday) is downloaded and converted to Parquet under ./data/meteoswiss/.

Collections

MeteoSwiss organises its open data into five categories. Category B (atmosphere measurements — radio soundings, ceilometer, ozone, etc.) is not yet released (B1 radio soundings expected first half of 2026).

A — Ground-based measurements

Station-level time series in CSV, split into time slices (historical, recent, now). Converted to Parquet.

C — Climate data

D — Radar data

Radar collections are large and require --grids to download.

E — Forecast data

GRIB2 forecast collections are large and require --grids to download.

Hail hazard maps

Static spatial reference grids showing expected hail grain size (cm) at different return periods. These are not categorised under A–E because they are static hazard assessments, not measured or forecasted time series — they represent probabilistic climatological analyses published as fixed reference maps.

Time slices

MeteoSwiss splits CSV data into three time slices, encoded in the filename:

Some collections (phenology, totaliser, yearly aggregates) don't use time slices — they publish a single file per station.

All timestamps are UTC. For 10-min and hourly data the timestamp marks the end of the interval (16:00 = 15:50:01–16:00:00). For daily, monthly, and yearly data the timestamp marks the start (2023-06-01 = the whole of June).

Installation

From PyPI:

pip install foehn

From source:

git clone https://github.com/kayhendriksen/foehn
cd foehn
pip install -e .

With Databricks extras (PySpark + Delta):

pip install "foehn[databricks]"

Requires Python ≥ 3.10.

Python API

Use foehn directly from notebooks or scripts:

import foehn

# List all available datasets
foehn.list_datasets()
# [{'dataset': 'smn', 'collection_id': 'ch.meteoschweiz.ogd-smn', 'category': 'A',
#   'subcategory': 'A1', 'description': 'Automatic weather stations',
#   'format': 'CSV', 'frequencies': ['t', 'h', 'd', 'm'],
#   'time_slices': ['historical', 'recent', 'now']}, ...]

# Load data directly into a Polars DataFrame (nothing written to disk)
df = foehn.load("smn", station="BER", frequency="d")

# Filter by multiple stations and frequencies
df = foehn.load("smn", station=["BER", "ZUR"], frequency=["d", "h"])

# Include historical data
df = foehn.load("smn", station="BER", frequency="d", time_slice=["historical", "recent"])

# Explore dataset metadata (fetched live from the API)
foehn.parameters("smn")   # column name mappings: shortname, description, unit, type, ...
foehn.stations("smn")     # station info: abbr, name, canton, altitude, lat, lon, ...
foehn.inventory("smn")    # what each station measures and since when

# Download a single dataset to disk
foehn.download("smn", data_dir="./data/meteoswiss")

# Download with specific time slices
foehn.download("smn", time_slice=["historical", "recent"])

# Convert downloaded CSVs to Parquet
foehn.to_parquet("smn", data_dir="./data/meteoswiss")

MCP server

foehn includes an MCP server that gives LLMs live access to MeteoSwiss data. It's published on the MCP Registry.

pip install "foehn[mcp]"

Add it to your MCP client config:

{
  "mcpServers": {
    "foehn": {
      "command": "foehn",
      "args": ["mcp"]
    }
  }
}

Tools available:

CLI reference

The CLI uses subcommands that mirror the Python API:

foehn list

List all available datasets.

foehn list

foehn download [DATASET...]

Download datasets. Without arguments, downloads all CSV collections. Specify one or more datasets to download specific ones.

foehn download              # all CSV collections
foehn download smn pollen   # specific datasets only

foehn to-parquet [DATASET...]

Convert downloaded CSVs to Parquet. Without arguments, converts all collections.

foehn to-parquet            # all collections
foehn to-parquet smn        # single dataset

foehn metadata KIND DATASET

Show dataset metadata fetched live from the API. KIND is one of parameters, stations, or inventory.

foehn metadata parameters smn   # what each column name means
foehn metadata stations smn     # station locations and info
foehn metadata inventory smn    # which station has which parameter

foehn load DATASET

Load a dataset from the API and print a preview (no files written to disk).

foehn load smn --station BER --frequency d
foehn load smn --station BER ZUR --frequency d h -n 50

Parquet files land in <data-dir>/parquet/<collection>/.

Environment variables

Settings can also be configured via environment variables. CLI flags always take precedence.

Databricks pipeline

The recommended setup uses Declarative Automation Bundles.

1. Set variables:

export BUNDLE_VAR_host=https://adb-xxx.azuredatabricks.net
export BUNDLE_VAR_alert_email=you@example.com

2. Deploy:

pip install databricks-cli
databricks bundle validate
databricks bundle deploy -t prod

This deploys two jobs:

• foehn_daily — runs at 13:30 UTC every day; downloads recent data and refreshes Delta tables
• foehn_historical — paused by default; trigger manually for first run or on Jan 1 for the annual archive slice

Data sources

License

MIT