forecast-mcp 0.1.0

An MCP server that gives any AI agent time-series forecasting superpowers (TimesFM 2.5 + statistical baselines).

forecast-mcp

Give any AI agent time-series forecasting superpowers.

An MCP server that lets Claude Code, Claude Desktop, Cursor, or any MCP client forecast a series of numbers — sales, traffic, usage, costs — and reason about the result. Powered by Google's TimesFM 2.5 foundation model, with a zero-dependency statistical baseline so it works the moment you install it.

Why

LLM agents can read, write, and run code — but they can't see the future. This gives them a clean forecast tool. The agent calls it, gets point forecasts + uncertainty bands + a compact trend/seasonality summary, and writes the explanation and recommendation itself.

Quickstart (30 seconds)

uvx forecast-mcp        # runs over stdio for local agents

Add to your Claude Desktop / Claude Code config:

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

Then ask your agent: "Forecast the next 6 months from this revenue data and tell me what to expect."

Enable the foundation model

pip install "forecast-mcp[timesfm]"

The server auto-detects TimesFM and uses it; otherwise it runs the baseline.

Tools

Tool	What it does
forecast	Forecast a single series with optional uncertainty bands.
list_backends	Report which engine is active (timesfm / baseline).
backtest	Hold out the last N points and compare TimesFM vs baseline performance (MAE/MAPE).

License

Apache-2.0