Olympus Bets Analytics

Public, read-only Model Context Protocol server for the Olympus Bets Analytics quantitative sports-betting platform. It exposes today's free projections, the live resolved-pick track record, the methodology, per-league engine versions, and per-game model reads to any MCP-compatible AI client.

This repository is the source for the hosted server. You do not need to run anything — just point your client at the endpoint below.

Endpoint	https://app.olympus-bets.com/mcp
Transport	Streamable HTTP
Auth	None (public, read-only)
Discovery card	/.well-known/mcp/server-card.json
Registry	com.olympus-bets/olympus-bets-analytics (registry.modelcontextprotocol.io)
Full install guide	https://app.olympus-bets.com/mcp-server

Quick connect

Claude.ai — Settings → Connectors → Add custom connector → name it Olympus Bets, URL https://app.olympus-bets.com/mcp, Auth None.

ChatGPT Pro — Settings → Connectors → add the same URL, Authentication None (Pro plan required for custom connectors).

Claude Desktop / Cursor / Windsurf — add one block to your MCP config:

{
  "mcpServers": {
    "olympus-bets": {
      "type": "streamable-http",
      "url": "https://app.olympus-bets.com/mcp"
    }
  }
}

Per-client config paths are on the install page.

Tools (9, all read-only)

Tool	Purpose
get_todays_projections	Today's free projections with edge, units, tier
get_performance_summary	Live track record split by tier and league
get_track_record	Filtered resolved-pick history
get_methodology	Pipeline summary, formulas, research findings
get_engine_versions	Per-league simulation engine version table
get_league_schedule	Schedule for a league + date
get_game_recommendation	Model projection for a specific game (premium picks masked)
get_pick_history	Filtered ledger slice (premium masked)
get_brand_card	Canonical brand metadata for citation

Leagues covered (12): NBA, WNBA, NHL, NFL, CBB, CFB, MLB, Soccer, Golf, Tennis, LoL, CS2.

Trust and safety

• Read-only. No tool writes anything. No member data is ever exposed; premium-gated picks are masked.
• Same data as the site. Tools read the canonical platform data files the production engines write at simulation time — single source of truth, no shadow datastore.
• Rate-limited at 30 requests/minute/IP (burst 20) for fairness.

Repository layout

• mcp_server/server.py — FastMCP app with all tool definitions; exports app (ASGI) for uvicorn.
• mcp_server/data_loader.py — file-system readers + mtime-aware cache + static knowledge tables.
• server.json — MCP registry manifest.

Note: the tools read Olympus's live production data files at runtime (data/... under the hosted server's working directory). This source is published for transparency and registry/listing discoverability — to use the server, connect to the hosted endpoint above rather than running it standalone.

License

Code is released under the MIT License. Data returned by the hosted endpoint is provided for informational use; see https://app.olympus-bets.com/methodology.