@general-liquidity/sharpebench-mcp 0.0.3

package/README.md

@@ -0,0 +1,36 @@
+# @general-liquidity/sharpebench-mcp
+
+An **MCP server** that exposes the [SharpeBench](https://github.com/general-liquidity/sharpebench) luck-robust scoring kernel as agent-callable tools. Point Claude (or any MCP client) at it and it can deflate a Sharpe, check pass^k reliability, audit a briefing for bias, or price an option's tail-risk — all from the deterministic Rust kernel, no network.
+
+## 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` | Black-Scholes price + Greeks + tail-selling risk |
+| `canary` | Derive a do-not-train contamination tripwire |
+
+All tools are read-only and deterministic — safe to expose without sandboxing.
+
+## Use it
+
+Add to your MCP client config (e.g. Claude Desktop's `mcpServers`):
+
+```json
+{
+  "mcpServers": {
+    "sharpebench": {
+      "command": "npx",
+      "args": ["-y", "@general-liquidity/sharpebench-mcp"]
+    }
+  }
+}
+```
+
+## License
+
+MIT OR Apache-2.0, at your option.

package/dist/server.js

@@ -0,0 +1,53 @@
+#!/usr/bin/env node
+/**
+ * SharpeBench MCP server — exposes the luck-robust scoring kernel as
+ * Model-Context-Protocol tools, so Claude and other agents can call
+ * "deflate this Sharpe / check pass^k / audit this briefing" in their tool loop.
+ *
+ * Every tool is read-only and deterministic (the kernel has no I/O), so the
+ * server is safe to expose without sandboxing.
+ */
+import { fileURLToPath } from "node:url";
+import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
+import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
+import { z } from "zod";
+import * as sb from "@general-liquidity/sharpebench";
+/** Run a kernel call, returning its result as pretty JSON or a typed error result. */
+function run(fn) {
+    try {
+        return { content: [{ type: "text", text: JSON.stringify(fn(), null, 2) }] };
+    }
+    catch (e) {
+        const message = e instanceof Error ? e.message : String(e);
+        return { content: [{ type: "text", text: `error: ${message}` }], isError: true };
+    }
+}
+/** Build the SharpeBench MCP server with all kernel tools registered. */
+export function createServer() {
+    const server = new McpServer({ name: "sharpebench", version: "0.0.3" });
+    server.tool("score", "Rank a field of agent submissions on the luck-robust composite (deflated Sharpe + pass^k + process discipline). Raw return is reported but is never the rank key. Returns ranked CompositeScore[].", { submissions: z.array(z.any()), config: z.any().optional() }, async ({ submissions, config }) => run(() => sb.score(submissions, config)));
+    server.tool("score_agent", "Score a single submission → one CompositeScore (deflated Sharpe, pass^k verdict, process score, rolling worst-case Sharpe).", { submission: z.any(), config: z.any().optional() }, async ({ submission, config }) => run(() => sb.scoreAgent(submission, config)));
+    server.tool("self_audit", "Fire the known gaming attacks at the scorer and report whether each is demoted (the benchmark's anti-gaming proof). No input.", async () => run(() => sb.selfAudit()));
+    server.tool("audit_briefing", "Audit a shared briefing artifact for input-side salience bias: per-asset attention caps, required counterbalancing, no performance-sorted return tables.", { briefing: z.any(), policy: z.any().optional() }, async ({ briefing, policy }) => run(() => sb.auditBriefing(briefing, policy)));
+    server.tool("score_allocation", "Score a target-allocation weight-vector trajectory: weight validity + L1 turnover/churn.", { trajectory: z.any(), policy: z.any().optional() }, async ({ trajectory, policy }) => run(() => sb.scoreAllocation(trajectory, policy)));
+    server.tool("greeks", "Black-Scholes price + Greeks (delta/gamma/theta/vega/rho) + tail-selling (short-gamma/vega) classification for one option.", {
+        spot: z.number(),
+        strike: z.number(),
+        t_years: z.number(),
+        rate: z.number(),
+        vol: z.number(),
+        is_call: z.boolean(),
+    }, async (params) => run(() => sb.greeks(params)));
+    server.tool("canary", "Derive a deterministic do-not-train contamination tripwire token from seed material.", { seed: z.string() }, async ({ seed }) => run(() => sb.canary(seed)));
+    return server;
+}
+async function main() {
+    const server = createServer();
+    await server.connect(new StdioServerTransport());
+}
+if (process.argv[1] === fileURLToPath(import.meta.url)) {
+    main().catch((e) => {
+        console.error(e);
+        process.exit(1);
+    });
+}

package/package.json

@@ -0,0 +1,44 @@
+{
+  "name": "@general-liquidity/sharpebench-mcp",
+  "version": "0.0.3",
+  "description": "MCP server exposing the SharpeBench luck-robust scoring kernel as agent-callable tools (deflated Sharpe, pass^k, process discipline, briefing audit, options Greeks).",
+  "keywords": [
+    "mcp",
+    "model-context-protocol",
+    "trading",
+    "sharpebench",
+    "benchmark",
+    "ai-agents"
+  ],
+  "license": "MIT OR Apache-2.0",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/general-liquidity/sharpebench.git",
+    "directory": "npm/mcp"
+  },
+  "type": "module",
+  "bin": {
+    "sharpebench-mcp": "dist/server.js"
+  },
+  "files": [
+    "dist"
+  ],
+  "scripts": {
+    "build": "tsc",
+    "test": "node --test",
+    "start": "node dist/server.js",
+    "prepublishOnly": "tsc"
+  },
+  "dependencies": {
+    "@general-liquidity/sharpebench": "^0.0.3",
+    "@modelcontextprotocol/sdk": "^1.0.0",
+    "zod": "^3.23.0"
+  },
+  "devDependencies": {
+    "@types/node": "^20.0.0",
+    "typescript": "^5.6.0"
+  },
+  "engines": {
+    "node": ">=18"
+  }
+}