Skip to main content

geosphere-mcp-server

PyPI Python License: MIT

MCP server for weather: current conditions, hourly forecasts, storm outlooks, and air quality for Austria and the Alpine region, via the Model Context Protocol.

It serves high-resolution GeoSphere Austria data — the AROME numerical forecast, the INCA analysis/nowcast, the C-LAEF ensemble for precipitation probability, and the WRF-Chem air-quality forecast. There is no worldwide fallback: a point outside the AROME grid returns an out-of-coverage notice, so pair this with a global weather source if you need one. Every response states which datasets produced it.

Output is compact emoji-markdown with metric units — built for smart-home and voice-assistant LLM pipelines where a terse, readable answer beats a JSON blob. Weather conditions are derived from physical parameters and reported with the Home Assistant condition vocabulary (sunny, partlycloudy, rainy, snowy, …).

Coverage

Where get_current_weather get_hourly_forecast get_storm_outlook get_air_quality
Austria INCA + nowcast + AROME AROME (≤60 h) + C-LAEF probability AROME, CAPE gated by CIN WRF-Chem (3 km)
Alps (non-AT) AROME only AROME (≤60 h) + C-LAEF probability AROME, CAPE gated by CIN WRF-Chem (3 km)
Rest of world not served not served not served not served

Coverage is decided by the API, not by a bounding box you configure: GeoSphere answers HTTP 400 for a point outside the grid, and the server renders that as

⚠️ Outside coverage — this server only serves Austria and the Alpine region.

That line is deliberately distinct from the transient failures (⚠️ Timeout…, ⚠️ No weather data available): being outside coverage is a permanent property of the location, so retrying will never help. It names no grid on purpose — the forecast tools query AROME (2.5 km) and get_air_quality queries WRF-Chem (3 km), so a point can be inside one and outside the other.

There is no multi-day forecast. GeoSphere publishes nothing beyond AROME's ~60 h, so the longest answer this server can give is roughly two and a half days of hourly rows.

Installation

Pass decimal latitude/longitude to every tool. There is no geocoder in the server — the calling LLM geocodes city names to coordinates itself.

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "geosphere": {
      "command": "uvx",
      "args": ["geosphere-mcp-server"]
    }
  }
}

Claude Code

claude mcp add geosphere -- uvx geosphere-mcp-server

From source (development)

{
  "mcpServers": {
    "geosphere": {
      "command": "uvx",
      "args": ["--from", "/path/to/geosphere-mcp-server", "geosphere-mcp-server"]
    }
  }
}

Home Assistant

It is a standard stdio MCP server, so it runs anywhere an stdio MCP server can be hosted — including alongside Home Assistant's Assist pipeline (register it the same way as any other stdio MCP server). Give the voice agent the coordinates of the places it should answer for, or let it geocode names.

Tools

get_current_weather

Current conditions for a point, from INCA, the 15-minute nowcast, and AROME.

Parameter Type Default Description
latitude float required Decimal latitude (e.g. 48.2208)
longitude float required Decimal longitude (e.g. 16.3738)
# Current Weather at 48.2208, 16.3738

🌡️ Temperature: 24.7°C (feels like 24.2°C)
🌤️ Condition: partlycloudy
💧 Humidity: 52%
💨 Wind: 2.9 m/s from 135° (gusts 7 m/s)
🌧️ Precipitation (last hour): 0 mm
📊 Pressure: 1016 hPa
☁️ Cloud cover: 30%
🕐 Timezone: Europe/Vienna (CEST)
📡 Source: GeoSphere (INCA + nowcast + AROME) — observed 15:20

get_hourly_forecast

Hour-by-hour forecast from AROME, with C-LAEF precipitation probability, up to ~60 h.

Parameter Type Default Description
latitude float required Decimal latitude (e.g. 48.2208)
longitude float required Decimal longitude (e.g. 16.3738)
hours int 24 Forecast hours (clamped to 1–60)
start string now Optional ISO 8601 start (e.g. 2026-07-22T15:00); forecast begins at/after this instant
# 3-Hour Forecast for 48.2208, 16.3738

AROME model, reference 2026-07-22 12:00 CEST · Source: GeoSphere (AROME + C-LAEF ensemble)

Wed 2026-07-22
  15:00: 24.7°C — partlycloudy, wind 3 m/s
  16:00: 23.9°C — rainy, 1.2 mm (70% chance), wind 4 m/s
  17:00: 22.5°C — cloudy, wind 3 m/s

The hours are grouped under a day-divider header (%a %Y-%m-%d in the point's local timezone) that repeats whenever the local date changes, so a window crossing midnight stays unambiguous. Requesting more hours than the AROME horizon provides appends a note naming where the horizon ends — there is nothing further ahead to reach for. Dry hours omit the precipitation and probability parts.

get_storm_outlook

Severe-weather outlook: peak gusts and thunderstorm timing. Reports figures, never a severity verdict — what counts as dangerous is the caller's judgement.

Parameter Type Default Description
latitude float required Decimal latitude (e.g. 48.2208)
longitude float required Decimal longitude (e.g. 16.3738)
# Storm Outlook for 48.2208, 16.3738

AROME model, reference 2026-08-10 14:00 CEST · Source: GeoSphere (AROME)

💨 Max gust next 1 h: 12 m/s (at Mon 2026-08-10 17:00)
💨 Max gust next 12 h: 24 m/s (at Mon 2026-08-10 20:00)
⛈️ Thunderstorm expected next 1 h: no
⚡ Next thunderstorm: Mon 2026-08-10 20:00 (CAPE 1800 J/kg)
🌡️ Max CAPE next 12 h: 1800 J/kg
🕐 Timezone: Europe/Vienna (CEST)

Two behaviours are worth knowing before you build on this:

  • Horizons round up to whole hours. The window starts at the top of the current hour, so the "next 1 h" figure covers the hour already under way plus the next one, and can report an event up to ~2 h out. Compare the returned timestamps yourself if you need a strict 60-minute answer.
  • Next thunderstorm can be in the past, by up to 59 minutes, when the storm hour is the one already under way. That means a storm is in progress — clamp a negative lead time to zero rather than assuming the stamp is in the future.
  • An all-clear names its horizon (none in the next 54 h). AROME nominally runs ~60 h, but a stale or truncated run reaches less far, so the span is measured from the rows actually returned. It is not an all-clear beyond the stated span.

A thunderstorm hour is one whose derived condition is lightning/lightning-rainy, or one where CAPE ≥ 1000 J/kg with weak inhibition and precipitation is forecast — the second branch catches thundersnow and hours with missing cloud data, and requires precipitation so that a dry high-CAPE afternoon does not raise a signal. Thunderstorm expected and Next thunderstorm both report unknown rather than a confident answer when the forecast holds no usable hour.

get_air_quality

Pollutant concentrations now, plus the European Air Quality Index for today, tomorrow and in two days.

Parameter Type Default Description
latitude float required Decimal latitude (e.g. 48.2208)
longitude float required Decimal longitude (e.g. 16.3738)
# Air Quality at 48.2208, 16.3738

🏷️ European AQI: 2 (fair) today · 3 (moderate) tomorrow · 2 (fair) in 2 days
🌫️ Concentrations (16:00): NO₂ 18 µg/m³ · O₃ 92 µg/m³ · PM10 21 µg/m³ · PM2.5 12 µg/m³
🕐 Timezone: Europe/Vienna (CEST)
📡 Source: GeoSphere (WRF-Chem + daily AQI, 3 km)

The AQI is reported as its 1–6 EEA band (1 good … 6 extremely poor), which is what GeoSphere publishes — there is no underlying numeric index to show alongside it. These are model forecasts, not station measurements — expect them to track a nearby monitoring station without matching it.

A point inside the 3 km grid can still come back empty when a WRF-Chem run is stale or incomplete; the response then says No air-quality data available for this location and still names the source, which is a different answer from being out of coverage.

Data sources & attribution

  • GeoSphere Austria Dataset API — AROME forecast, INCA analysis/nowcast, C-LAEF ensemble, WRF-Chem air quality. Data licensed under CC-BY 4.0. © GeoSphere Austria.

The API is keyless and intended for non-commercial use. When you redistribute its data, keep the attribution.

Rate limits

The GeoSphere Dataset API allows 5 requests/second and 240 requests/hour. Each call issues a small burst of concurrent requests (three for current weather, two for hourly and air quality, one for the storm outlook); on an HTTP 429 the server retries once (when the API asks for a short wait) and otherwise returns a rate-limit notice.

Development

# Install dependencies (creates .venv from the locked versions)
uv sync

# Lint & format
ruff check .
ruff format .

# Run unit tests
pytest -m "not integration"

# Run integration tests (hits the live GeoSphere API)
pytest -m integration

License

MIT

Metadata

Release files for geosphere-mcp-server 0.5.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for geosphere-mcp-server 0.5.0
File Size Uploaded Provenance
geosphere_mcp_server-0.5.0.tar.gz 207.2 kB With provenance Details

Built distribution (wheel)

Table of built distributions (wheels) for geosphere-mcp-server 0.5.0
File Interpreter ABI Platform
geosphere_mcp_server-0.5.0-py3-none-any.whl Python 3 none any Details

Total release size: 250.6 kB

Release files / geosphere_mcp_server-0.5.0.tar.gz

Download URL geosphere_mcp_server-0.5.0.tar.gz
Size 207.2 kB
Tags Source
SHA-256 checksum
How to use checksums
ffa29ceafc8ea472791ac024417eaaed0ee1930f6d8d478eb4af39a9e9639acc
BLAKE2b-256 checksum
How to use checksums
71d0e34e084159c1e73487ca42d4ee387de8e4869e87c7c64c4999862d6f48d1
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 11, 2026.

Transparency log

Release files / geosphere_mcp_server-0.5.0-py3-none-any.whl

Download URL geosphere_mcp_server-0.5.0-py3-none-any.whl
Size 43.4 kB
Tags Python 3
SHA-256 checksum
How to use checksums
7c2c9a361edd3b06dc023d021262840dc802e047d895803c9dd1afa842ceaaed
BLAKE2b-256 checksum
How to use checksums
51c11e10db1852f7975d5a581d0990c6b73fafe2a6793b13ddbf1eaef5674ca6
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 11, 2026.

Transparency log

Security for geosphere-mcp-server 0.5.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

All 2 files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for geosphere_mcp_server-0.5.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 11, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page