If you're building an AI agent that needs to reason about live sports — fantasy advice, betting-adjacent research (Kalshi-style prediction markets, not sportsbook odds), or just "what happened to this player" — PlayCaller runs an MCP server so the agent can pull real data instead of guessing from training-data cutoffs.

What the server exposes

  • get_injury_signals — active beat-reporter injury and intelligence signals, filterable by sport and severity, refreshed every 2 hours
  • get_player_identity — resolves a player name or platform ID to a canonical identity across ESPN, Sleeper, Yahoo, and NFL.com
  • get_anomaly_scores — players whose usage patterns deviate from a 3-season baseline (0-100 score; above 70 is actionable)
  • get_prediction_markets — active Kalshi-related props with PlayCaller's AI confidence scores attached
  • get_player_projections — fantasy-point projections using PlayCaller's scoring engine plus injury and matchup signals

Wiring it up

The server speaks streamable HTTP at https://playcallerapp.com/mcp. A GET request returns the tool list with no auth required, for discovery; calling the tools requires the same X-PlayCaller-Key header as the REST API. In a Claude Desktop-style config:

{
  "mcpServers": {
    "playcaller": {
      "url": "https://playcallerapp.com/mcp",
      "headers": { "X-PlayCaller-Key": "<YOUR_API_KEY>" }
    }
  }
}

The package is also published on npm as @playcaller/playcaller-mcp if you'd rather install a local wrapper than point directly at the hosted endpoint.

Ready to build? Start a free 14-day Developer Sandbox — no credit card, API key in under a minute.