🟢 NVIDIA NIM Expert Orchestrator (MCP)

A powerful Model Context Protocol (MCP) server that orchestrates a swarm of NVIDIA NIM models. It features an intelligent consensus engine that consults specialized experts and synthesizes their responses for maximum accuracy and creativity.

🚀 Key Features

• 8 Specialized Expert Groups: Automated routing to models optimized for coding, UI/UX, security, reasoning, and more.
• Consensus & Synthesis: Optional multi-model polling to compare expert opinions and generate a unified, high-quality answer.
• Library & Server Modes: Use it as a standalone MCP server for Claude/Cursor or as a TypeScript library in your own apps.
• Production Ready: Built-in retries, timeouts, and fallback mechanisms using the latest Llama 3.3 70B for synthesis.

🛠️ Installation & Setup

1. Get your NVIDIA API Key

Get your free or enterprise API key at build.nvidia.com.

2. Install as a Server

To use this with a client like Claude Desktop, add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "nvidia-orchestrator": {
      "command": "npx",
      "args": ["-y", "mcp-nvidia-orchestrator"],
      "env": {
        "NVIDIA_API_KEY": "your_api_key_here"
      }
    }
  }
}

3. Use as a Library

npm install mcp-nvidia-orchestrator

import { NvidiaOrchestrator } from 'mcp-nvidia-orchestrator/orchestrator';

const orchestrator = new NvidiaOrchestrator("your_api_key");

const response = await orchestrator.ask(
  "Create a high-performance shader for a glass effect",
  "coding",
  true // Enable multi-model consensus
);

console.log(response.content);

🧠 Expert Groups

The orchestrator routes your requests to specialized model groups:

| Category | Best For... | Primary Model | | :--- | :--- | :--- | | Reasoning | Logic, Math, Strategy | Llama-3.1-405b-instruct | | Coding | Software engineering, Code review | DeepSeek-Coder-V2-Instruct | | UI/UX | Aesthetics, Premium layouts, Animations | Llama-3.1-70b-instruct | | Vision | Mockups, Diagrams, Bug screenshots | Llama-3.2-90b-vision-instruct | | Creative | Copywriting, Social Media, Assets | Nemotron-340b-instruct | | Security | PII detection, Vulnerabilities | Llama-3-70b-instruct | | Data | Extraction, Embeddings, SQL | Llama-3.1-8b-instruct | | Ops | Infrastructure, K8s, Performance | Llama-3.1-70b-instruct |

🏗️ Architecture

The orchestrator uses a Consensus Engine:

1. Selection: Picks the best expert group for the task.
2. Parallel Execution: Queries 2-3 expert models simultaneously.
3. Synthesis: A "Master Model" (Llama 3.3 70B) analyzes all responses and crafts the definitive final output.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Developed with ❤️ for the AI Community.