swiss-outdoor-mcp

An MCP server that gives an LLM client reliable, computed facts for planning Swiss mountain outings by public transport. The v0.1 use case is paragliding:

"Saturday, from Lausanne, where can I go paragliding by train?"

The server does not answer that question. It provides the deterministic pieces — launch sites, ensemble-based flyability probabilities, train connections — and the LLM composes them. There are no LLM calls inside the server.

Status

Work in progress towards v0.1; the current pre-release is 0.1.0a1. Working today: list_sites, get_flyability and get_connections. Still to come: estimate_trip_co2, offline fixture mode and the eval harness. See docs/SPEC.md.

Related MCP server: mcp-camptocamp

Example

A real get_flyability call for the Fiesch launch (Valais, faces SE–SW, 2163 m), made on 2026-09-25 for the next day. The hourly list is trimmed to its first and last hours; the real response has all eight, 10:00 to 17:00.

{
  "site_id": "fiesch",
  "date": "2026-09-26",
  "p_flyable": 1.0,
  "hourly": [
    {
      "time": "2026-09-26T10:00:00+02:00",
      "p_wind_ok": 1.0,
      "p_gust_ok": 1.0,
      "p_dry": 1.0,
      "p_direction_ok": 1.0,
      "p_flyable": 1.0,
      "wind_kmh": { "median": 1.8, "p10": 1.6, "p90": 2.5 }
    },
    {
      "time": "2026-09-26T17:00:00+02:00",
      "p_wind_ok": 1.0,
      "p_gust_ok": 1.0,
      "p_dry": 1.0,
      "p_direction_ok": 1.0,
      "p_flyable": 1.0,
      "wind_kmh": { "median": 8.6, "p10": 7.4, "p90": 10.8 }
    }
  ],
  "model": "icon_d2_eps",
  "n_members": 20,
  "grid_elevation_m": 2173.0,
  "criteria": {
    "max_wind_kmh": 20.0,
    "max_gust_kmh": 30.0,
    "max_precip_mm_h": 0.1,
    "direction_tolerance_deg": 45.0,
    "window_start": "10:00:00",
    "window_end": "17:00:00",
    "min_consecutive_hours": 3
  },
  "method": "Each of the 20 members of the icon_d2_eps ensemble is checked hour by hour against the criteria (all limits inclusive; wind 10 m above the model's terrain). Hourly p_* values are the share of members meeting each criterion. The day-level p_flyable is the share of members with at least 3 consecutive flyable hours between 10:00 and 17:00 Europe/Zurich, both ends included.",
  "generated_at": "2026-09-25T16:19:01+02:00",
  "disclaimer": "Indicator only, computed from a weather-model ensemble. It does not replace pilot judgment, an on-site assessment of conditions, or official aviation weather forecasts.",
  "attribution": "Weather data by Open-Meteo.com (https://open-meteo.com/), CC BY 4.0."
}

How to read it:

• p_flyable: 1.0 means all 20 ensemble members have at least 3 consecutive hours inside the criteria. It is a share of members, not a calibrated probability.

• model says which ensemble answered: the 2 km icon_d2_eps (20 members) for today and tomorrow, the 13 km icon_eu_eps (40 members) further out.

• grid_elevation_m is the altitude of the model cell. Here it is 10 m from the real launch; on other sites the gap can be hundreds of metres.

• The LLM client is expected to pass disclaimer and attribution on to the user.

Install and run

uv sync
uv run swiss-outdoor-mcp     # speaks MCP over stdio

Development

uv run ruff check . && uv run ruff format --check . && uv run mypy src && uv run pytest

See CONTRIBUTING.md. Verified facts about every upstream API live in docs/api-notes.md — read it before touching a client.

Limitations

This is not a flight-safety tool. Flyability output is an indicator built from a public weather model; it does not replace pilot judgment, a site briefing, or official aviation weather.

• p_flyable is the share of ensemble members that agree, not a calibrated probability.

• Winds are 10 m above the model's terrain, not at launch height. The model cell can sit hundreds of metres from the real launch altitude; grid_elevation_m reports it. No thermals, no foehn, no upper winds.

• Near-calm hours can score low: the direction check applies whatever the wind speed, so a 1 km/h breeze from the wrong side fails it.

• The forecast reaches about four days out. The exact horizon moves with each model run and is read from every response.

• The sites' access_notes (the last leg from the stop to the launch) are not filled in yet.

Data sources

• Weather: Open-Meteo Ensemble API — weather data by Open-Meteo.com, CC BY 4.0.

• Public transport: transport.opendata.ch (an unofficial API over the Swiss timetable; its terms of use are not stated — see docs/api-notes.md).

Licence

MIT — see LICENSE.