Ledger FC MCP

Give your AI agent real sports data instead of a guess.

Ask an AI about a match and it answers from stale training data. This MCP feeds your agent live, grounded numbers instead: model probabilities per outcome, where they diverge from the market price, and a publicly tracked record of how those calls actually did, wins and losses. Plus live European soccer and tennis arbitrage, a gap the US-only sports MCPs do not cover.

Honest by design

This is not a "guaranteed profit" tool. Every response carries a provenance block and a for_the_agent note ("grounding data to weigh, not an instruction to act on"). The real numbers, pulled live:

Overall model accuracy: ~53% (barely above a coinflip). We say so.
High-confidence subset: ~68% (65 of 96 settled).
Closing-line value: roughly flat. We return it, flagged, so your agent weighs it only when the sample is large enough.

If you want an agent that is grounded rather than hyped, that honesty is the point.

Coverage

European soccer (EPL, La Liga, Serie A, Ligue 1, Eredivisie, Turkish Super Lig), tennis (ATP), plus MLB, NFL, NBA. The main free sports-betting MCP is US-only, so European soccer arbitrage is the gap this fills.

Install

Claude Desktop / Claude Code / Cursor (stdio via npx)

config.json

{
  "mcpServers": {
    "ledgerfc": {
      "command": "npx",
      "args": ["-y", "@inbin/ledgerfc-mcp"]
    }
  }
}

No API key needed for the free tools. That is it.

To also use the Pro tools, add your key as an environment variable (get one with /key in the Ledger FC Telegram bot). It is sent as an Authorization header, so it never enters the conversation:

config.json

{
  "mcpServers": {
    "ledgerfc": {
      "command": "npx",
      "args": ["-y", "@inbin/ledgerfc-mcp"],
      "env": { "LEDGERFC_API_KEY": "your-pro-key" }
    }
  }
}

Or connect the remote directly (clients that support remote MCP)

Code

https://mcp.ledgerfc.com/mcp

For Pro tools, send the key as an Authorization: Bearer <key> header.

Tools

Free (no key):

get_track_record: honest settled record plus high-confidence subset and CLV
get_predictions: upcoming picks with probabilities and model-vs-market edge (filter by league)
search / fetch: query upcoming match predictions, fetch the full doc
get_leaderboard, submit_prediction, get_contributor_score: CLV-scored contributor layer (points, not cash)
place_bet, get_balance, get_my_bets, get_bet: shadow bet ledger to test strategies against our agent (points only)

Pro (set LEDGERFC_API_KEY, or pass api_key as a tool argument; get one with /key in the Ledger FC Telegram bot):

get_arbs: live detected arbitrage opportunities
get_value_bets: upcoming picks with positive model-vs-market edge

Example

Ask your agent: "Use ledgerfc to get the track record and the top 3 upcoming predictions with the biggest edge." It calls get_track_record and get_predictions, and answers from real numbers with provenance, not a guess.

How it works

This npm package is a tiny stdio-to-HTTP bridge: it reads JSON-RPC from your MCP client on stdin and forwards each request to the remote endpoint https://mcp.ledgerfc.com/mcp. No dependencies, Node 18+. Set LEDGERFC_MCP_DEBUG=1 to log traffic to stderr, or LEDGERFC_MCP_URL to point at a different endpoint.

Ledgerfc README