MCP Reference¶

Weather Signal exposes the same weather workflows through MCP for agents that prefer tools over shell commands.

For client configuration examples, see MCP Clients.

Stdio¶

weather-signal server start --transport stdio

Example client config:

{
  "mcpServers": {
    "weather-signal": {
      "command": "weather-signal",
      "args": ["server", "start", "--transport", "stdio"]
    }
  }
}

Streamable HTTP¶

weather-signal server start \
  --transport streamable-http \
  --http-host 127.0.0.1 \
  --http-port 8768 \
  --http-path /mcp

Health probes:

curl http://127.0.0.1:8768/healthz
curl http://127.0.0.1:8768/readyz

Keep HTTP bound to loopback unless an authenticating proxy controls access. The server prints a warning when --http-host is not 127.0.0.1, localhost, or ::1.

Streamable HTTP is stateless by default. Use --http-stateful-mode only for trusted local clients because stateful sessions are held in process memory for the lifetime of the server process.

Tools¶

Tool	Use
`geocode`	Resolve ambiguous place names
`current_weather`	Fetch current weather fields
`daily_forecast`	Fetch daily forecast variables
`hourly_forecast`	Fetch hourly forecast variables
`demand_signal`	Return daily demand-oriented weather features
`weather_summary`	Return compact counts and headline over the forecast window
`threshold_days`	Filter days by rain, precipitation, heat, cold, or wind rules
`historical_weather`	Fetch archive data for backtesting
`list_places`	List saved aliases
`cache_status`	Inspect local response cache state

Prefer weather_summary first for broad business context, then call demand_signal or threshold_days when the agent needs row-level features or decision-rule matches.