Skip to content

Skforecast AI

Python PyPI Project Status: Active PyPI Downloads License Documentation GitHub MCP server

Skforecast AI is an AI-assisted forecasting package from the skforecast team. It combines a deterministic forecasting engine powered by skforecast with an optional LLM reasoning layer.

Provide a time series and the assistant can profile the data, choose a forecasting strategy using established best practices, evaluate its performance, and return both the forecast and the runnable skforecast code that produced it.

Why Skforecast AI?

  • 🎯 Deterministic by design: The rule-based forecasting engine produces consistent results for the same input.
  • 🔍 Inspectable and reproducible: The generated script is the code that ran, so you can inspect, version, and execute it independently with skforecast.
  • ⚡ From data to forecast in one call: Automates profiling, model and estimator selection, feature engineering, and backtesting.
  • 💻 Python and CLI workflows: Use the assistant from Python or run the complete pipeline from the terminal.
  • 💬 Optional LLM reasoning: Get plain-language explanations and configuration advice while keeping the core forecasting workflow available offline.
  • 🏗 Built on skforecast: Supports recursive and direct forecasters, multi-series forecasting, statistical models, and foundation models.

Installation

Skforecast AI requires Python 3.10 or later.

pip install skforecast-ai

Install the optional LLM reasoning layer with:

pip install "skforecast-ai[llm]"

Quick Start

from skforecast.datasets import load_demo_dataset
from skforecast_ai import ForecastingAssistant

data = load_demo_dataset(verbose=False)
assistant = ForecastingAssistant()
result = assistant.forecast(data=data, target="y", steps=12)

print(result.predictions)
print(result.metrics)
print(result.code)

The result includes the predictions, backtesting metrics, data profile, selected modeling plan, and the standalone skforecast script used to produce the forecast.

Use it from your coding agent

Since version 0.4.0, Skforecast AI includes an MCP server that brings skforecast to coding agents such as Claude Code, Claude Desktop or VS Code. Ask in plain language ("Forecast the next 12 months of data/sales.csv and tell me how accurate it is") and the agent profiles the file, plans a forecaster, backtests it, compares candidates and forecasts.

Install it as a plugin. Run these commands inside Claude Code:

/plugin marketplace add skforecast/skforecast-ai
/plugin install skforecast-ai@skforecast-ai

Register this command as an MCP server in your client (requires uv):

uvx skforecast-ai-mcp --allow-dir /absolute/path/to/project

The server is listed in the official MCP registry as io.github.skforecast/skforecast-ai.

  • The rules decide, the agent explains: the forecaster, estimator, lags, metric and cross-validation come from deterministic rules, not from the language model. The agent provides the model and explains the results.
  • Every result comes with its code: each result includes the skforecast script that produced it, which you can run on its own.
  • Your data stays local: the server only reads CSV files inside the allowed directory and never returns data rows to the agent, only summaries.

The server requires skforecast 0.26.0 or later. See the MCP server user guide for the configuration of each client, the available tools and the security model.

Learn More

Feedback and Issues

Skforecast AI is developed by the skforecast team. If you encounter a problem or have a suggestion, please open an issue.