cmer81/open-meteo-mcp/open-meteo-advanced

Open-Meteo MCP — Advanced

Overview

Use this skill when:

• The user asks for a specific weather model (ECMWF, GFS, DWD ICON, Météo-France, JMA, MET Norway, GEM)
• The user wants to compare models (make parallel calls, one per model tool)
• Forecasts beyond 16 days are needed
• Ensemble uncertainty / confidence intervals are requested
• A seasonal outlook (1–7 months) or climate projection (to 2050) is needed

For everyday weather questions, use open-meteo instead.

Model Selection Guide

Geographic constraints:

• dwd_icon_forecast: Europe only — do not use for other regions.
• metno_forecast with metno_nordic: Nordic region only. metno_seamless is Nordic-primary with global extension via blending, not a general-purpose global model.
• meteofrance_forecast: France + DOM-TOM + Mediterranean/Europe coverage.

Key Parameters

Model-specific forecast tools

Applies to: dwd_icon_forecast, gfs_forecast, meteofrance_forecast, ecmwf_forecast, jma_forecast, metno_forecast, gem_forecast

All share the same parameters as weather_forecast (see open-meteo skill), plus:

\`metno_forecast` can omit `models` to use the default Met.no model. \\*At least one variable group required.

Multi-model comparison: Make parallel tool calls, one per model.

Model key examples:

ECMWF warning: model IDs with an underscore before the resolution digits (ecmwf_ifs_025, ecmwf_aifs_025_single) and ecmwf_ifs_hres_9km are NOT valid on ecmwf_forecast. Use ecmwf_ifs025 / ecmwf_aifs025_single / ecmwf_ifs.

ensemble_forecast

Response format: Each variable is returned as one array per ensemble member, suffixed with the model name when models contains more than one entry:

{
  "hourly": {
    "time": ["2024-01-01T00:00", "2024-01-01T01:00", "..."],
    "temperature_2m_member01": [2.1, 2.3, "..."],
    "temperature_2m_member02": [1.8, 2.0, "..."],
    "temperature_2m_member03": [2.4, 2.6, "..."]
  }
}

To derive uncertainty ranges, calculate min/max/percentiles across all _memberNN arrays for each timestep.

seasonal_forecast

\*At least one of hourly, daily, weekly, or monthly is required.

Output represents ensemble anomalies relative to climatology, not absolute forecasts.

climate_projection

Data note: Dates before the current year represent CMIP6 model simulation output for validation purposes, not observed historical measurements. For real historical weather data, use weather_archive.

Examples

"Get the ECMWF forecast for Bordeaux tomorrow"

1. geocoding with name: "Bordeaux" → coordinates
2. ecmwf_forecast with coordinates + models: "ecmwf_ifs", daily: ["temperature_2m_max", "temperature_2m_min", "precipitation_sum", "weather_code"], forecast_days: 2, timezone: "auto"

"Compare DWD ICON and GFS forecasts for Berlin this week"

1. geocoding with name: "Berlin" → coordinates
2. Two parallel calls:

• dwd_icon_forecast with models: "dwd_icon_seamless", daily: ["temperature_2m_max", "precipitation_sum"], forecast_days: 7, timezone: "auto"
• gfs_forecast with models: "ncep_gfs_global", daily: ["temperature_2m_max", "precipitation_sum"], forecast_days: 7, timezone: "auto"

"Show me forecast uncertainty for Paris next week"

1. geocoding with name: "Paris" → coordinates
2. ensemble_forecast with coordinates + models: "ecmwf_ifs025_ensemble", hourly: ["temperature_2m"], forecast_days: 10

The response will contain temperature_2m_member01, temperature_2m_member02, etc. — calculate spread across members for uncertainty.

"What will the climate be like in Lyon in 2040?"

1. geocoding with name: "Lyon" → coordinates
2. climate_projection with coordinates + start_date: "2040-01-01", end_date: "2040-12-31", models: ["MRI_AGCM3_2_S"], daily: ["temperature_2m_max", "temperature_2m_min", "precipitation_sum"]

Best Practices

• Don't use regional model tools outside their geographic coverage — dwd_icon_forecast for Asia will return empty or incorrect data.
• `ecmwf_forecast` accepts only the ECMWF model IDs listed above (IFS, AIFS and the Europe ensemble means). Any other key causes HTTP 400.
• Ensemble output is member arrays, not scalar values — process all _memberNN keys to derive uncertainty ranges.
• Climate data before the current year is CMIP6 simulation, not observed data — use weather_archive for real historical measurements.
• `seasonal_forecast` outputs anomalies, not absolute values — it answers "warmer than usual?" not "what temperature exactly?".
• Geocode first if you only have a city name.