@general-liquidity/sharpebench-mcp

An MCP server that exposes SharpeBench's luck-robust quantitative evaluation as agent-callable tools. Point an MCP client at it to deflate a Sharpe, check pass^k reliability, compare regimes, audit a briefing, or price option tail risk. Every result comes from the deterministic Rust kernel with no network access.

Tools

| Tool | What it does | |---|---| | score | Rank a field of submissions on the luck-robust composite | | score_agent | Score one submission → deflated Sharpe / pass^k / process / rolling Sharpe | | self_audit | Fire known gaming attacks at the scorer (anti-gaming proof) | | audit_briefing | Audit a shared briefing for input-side salience bias | | score_allocation | Score a weight-vector trajectory (validity + turnover) | | greeks | One long European option: price, Greeks and local exposure flags | | canary | Derive a do-not-train contamination tripwire | | is_my_sharpe_real | Deflate one return series for its search footprint and render the honesty verdict (trials_sr_std and sr_benchmark annualized, periods_per_year default 252) | | regime_compare | Compare aligned returns inside caller-supplied regimes | | percentile_selection | Compare the observed point winner with a bootstrap-percentile winner | | decompose_uncertainty | Report aleatoric, epistemic, and distributional diagnostic legs | | crowding_half_life | Evaluate a caller-calibrated crowding-decay prior | | classify_disqualification | Name every hard-gate and advisory signal that fired |

All tools are read-only and deterministic. They do not execute entrant code or access the network.

Use it

Add to your MCP client config (e.g. Claude Desktop's mcpServers):

{
  "mcpServers": {
    "sharpebench": {
      "command": "npx",
      "args": ["-y", "@general-liquidity/sharpebench-mcp"]
    }
  }
}

License

MIT OR Apache-2.0, at your option.