ameva-mcp-hub

Universal Polyglot WebAssembly & AI Vector Model Context Protocol (MCP) Hub
Dynamic Multi-Repository Subscription, Zero-Install WASI Runtime, and Heuristic Execution Engine.

• NPM Package: https://www.npmjs.com/package/ameva-mcp-hub
• Official Documentation Portal: https://uno-km.vercel.app/lib/mcp/
• GitHub Repository: https://github.com/uno-km/ameva-mcp-hub

1. Executive Summary

ameva-mcp-hub is a high-performance, zero-browser native execution hub and Model Context Protocol (MCP) server. It enables AI clients (Claude Desktop, Cursor, Zed) to dynamically ingest, execute, and scale multi-language tools (C++, Rust, Java, Python, Go, JavaScript) from remote GitHub repositories inside a single lightweight Node.js process without requiring local compilers or runtimes (No JDK, No Rustc, No Clang required on host).

2. Key Architectural Invariants

• Polyglot WebAssembly In-Memory Core: Executes compiled .wasm binaries across C++, Rust (wasm32-wasip1), Java (TeaVM / GraalVM), and Go (TinyGo) with in-memory execution latency under 1ms.
• Zero-Browser Pure Node.js Architecture: Eliminates browser and Puppeteer overhead, booting in ~450ms with a ~30MB memory footprint.
• Dynamic Multi-Repository Subscription: Allows users to register and manage multiple GitHub tool repositories on the fly via HTTP API or Web Dashboard, automatically broadcasting notifications/tools/list_changed to connected AI agents without server restarts.
• AI Vector & Semantic Ranking Engine: Native high-dimensional Cosine Similarity (vector_cosine_similarity) and Top-K semantic ranking (vector_top_k_search) for localized RAG pipelines.
• Heuristic Adaptive Execution Router (smart_exec): Evaluates workload risk and host binary availability to route tasks automatically between HOST_NATIVE (for workspace builds/git) and WASM_SANDBOX (for isolated math/vectors/untrusted code).

3. Quick Start (Zero-Install CLI)

3.1 Direct Execution via NPX

npx ameva-mcp-hub

3.2 Claude Desktop Configuration

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "ameva-mcp-hub": {
      "command": "npx",
      "args": ["-y", "ameva-mcp-hub"]
    }
  }
}

4. SDK Library Usage

ameva-mcp-hub can be embedded directly into custom Node.js and TypeScript backends as an isolated execution engine:

npm install ameva-mcp-hub

import { 
  WasiRunner, 
  NodeExecutor, 
  heuristicRouter, 
  BridgeGateway 
} from "ameva-mcp-hub";

// 1. Direct In-Memory WASM Execution (Zero-JVM / Zero-Toolchain)
const runner = new WasiRunner({ args: ["100", "20"] });
const res = await runner.run(wasmBuffer);
console.log("Stdout:", res.stdout);

// 2. AI Vector Cosine Similarity
const executor = new NodeExecutor();
const similarity = await executor.executeTool("vector_cosine_similarity", {
  vecA: [0.1, 0.5, 0.9],
  vecB: [0.1, 0.5, 0.88]
});
console.log(similarity.content[0].text);

// 3. Heuristic Route Evaluation
const route = heuristicRouter.evaluate({ command: "git status" });
console.log(route.target); // "HOST_NATIVE"

5. Verification & Test Suites

The codebase includes deterministic, comprehensive automated test suites:

# Run All Master Test Suites
npm run test:all

# Individual Suites
npm run test:subscription  # Multi-Repo Real-Time Tool Propagation
npm run test:universal     # C++, Rust, Java WASM & AI Vector Tests
npm run test:heuristic     # Heuristic Adaptive Execution Router
npm run test:pure-node     # Zero-Browser Pure Node.js Stdio Test
npm run test:regression    # Config Hot-Reload & Parallel Host IPC
npm run lint:neutral       # Open-Source Neutrality & Compliance

6. Official Documentation & Portal

• Official Portal: https://uno-km.vercel.app/AMEVA-MCP-Hub/
• Issue Tracker: https://github.com/uno-km/ameva-mcp-hub/issues

7. License

Distributed under the MIT License.