The explainability tool is what separates this from a plain prediction endpoint: you can ask which appliance drove a forecast and over what horizon, then line that against the actual price for the period. Note the licence before you build on it — the author restricts use to academic evaluation, reference and non-commercial educational purposes with attribution, and asks that commercial use be arranged directly.
A Python MCP server sitting on top of a domestic energy-consumption prediction system: ensemble machine-learning models for the forecast, SHAP for explainability, and the ESIOS API for electricity prices.
- `predict_consumption()` — consumption forecasts between an `init_date` and an `end_date`
- `get_consumption_analysis()` — historical analysis over a date range, broken down by appliance
- `explain_predictions()` — SHAP values for a given `appliance` over a `horizon`, so the forecast is not a black box
- `get_precio()` — electricity prices for a date range and a `price_type`
- `get_precio_inteligente()` — price for a `target_date` with automatic fallback when the primary source has nothing
- Personalised optimisation recommendations built on the analysis, rather than generic advice
Python 3.8 or higher, an ESIOS account with Red Eléctrica de España, and the token in `ESIOS_API_TOKEN`. The client launches it with uv from the project directory.
Build from source — clone the repository and build it, then point your client at the binary