Weather Edge MCP Server

Calibrated weather probability signals for Kalshi prediction markets.

The only MCP server for weather prediction market intelligence. Uses dual-model forecasting (NWS + GFS ensemble) to find mispriced temperature markets.

Install

pip install weather-edge-mcp

Quick Start

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "weather-edge": {
      "command": "python",
      "args": ["-m", "weather_edge_mcp"]
    }
  }
}

Then ask Claude: "What are today's best Kalshi weather edge opportunities?"

Cursor / Windsurf

Add to your MCP settings with the same command: python -m weather_edge_mcp

Command Line

# stdio mode (for AI tools)
weather-edge-mcp

# SSE mode (for web clients)
weather-edge-mcp --transport sse --port 8050

Tools

Tool	Description
get_weather_signals(city)	Edge signals for a city's Kalshi weather markets — blended probability vs market price, confidence level, expected value
get_all_signals()	Full scan across all 5 cities, sorted by confidence
get_forecast(city)	NWS forecast (bias-corrected) + GFS 31-member ensemble distribution
get_station_observation(city)	Real-time METAR from the exact ASOS station Kalshi settles on
list_cities()	Available cities with calibration parameters

Cities: nyc, chicago, denver, miami, la

How It Works

Kalshi weather markets settle on NWS Climate Reports from specific ASOS stations. Most traders use the raw NWS forecast, but it has systematic biases:

• Miami: NWS gridpoint overshoots MIA Airport by ~3°F
• NYC: NWS gridpoint overshoots Central Park by ~1°F
• Denver: Mountain terrain makes forecasts ~2-4°F less reliable

Weather Edge corrects for these biases and combines two independent models:

Model 1: Calibrated NWS Gaussian

Per-city bias correction + calibrated sigma (uncertainty). Coastal cities (sigma=3°F) vs mountain (sigma=4°F).

Model 2: GFS 31-Member Ensemble

31 different forecast runs from Open-Meteo's free API. Gives real probability distributions, not assumptions.

Blended Model

Adaptive weighted average. When ensemble members agree (low spread), ensemble gets more weight. When they disagree, NWS stabilizes.

Confidence levels:

• HIGH — both models agree within 10%, net EV > 5 cents
• MODERATE — models agree within 20%
• LOW — models diverge but one side shows edge

Example Output

> get_weather_signals("chicago")

# Weather Edge — Chicago (Midway)
NWS forecast: 77°F (adjusted: 76°F) — Mostly Cloudy

## HIGH Confidence
- **BUY YES 76° to 77°** | NWS: 26.1% | Ensemble: 25.8% |
  Blended: 25.9% | Market: $0.01 | Edge: +25.4% | EV: $+0.248
- **BUY NO 78° to 79°** | NWS: 3.8% | Ensemble: 0.0% |
  Blended: 1.3% | Market: $0.73 | Edge: +25.2% | EV: $+0.239

## MODERATE Confidence
- **BUY YES 75° or below** | NWS: 84.1% | Ensemble: 100.0% |
  Blended: 94.7% | Market: $0.11 | Edge: +83.2% | EV: $+0.820

12 positive EV signals found

Data Sources (all free, no API keys)

Source	What	URL
NWS Weather API	Gridpoint forecasts	api.weather.gov
Open-Meteo	GFS 31-member ensemble	open-meteo.com
Aviation Weather	Real-time METAR observations	aviationweather.gov
Kalshi Trade API	Market pricing (public)	api.elections.kalshi.com

Why This Exists

I kept losing money on Kalshi weather markets because the raw NWS forecast is systematically wrong for some cities. This MCP server corrects for those biases and gives AI agents the calibrated probabilities they need to identify genuine mispricings.

License

MIT