nvidia-mcp

The only standalone NVIDIA NIM wrapper for MCP. 20 AI tools across 189 models -- code review, security scanning, RAG, content safety, translation, OCR, video analysis, and more. All powered by NVIDIA's free NIM API tier.

No competitor exists. NVIDIA's own NeMo toolkit requires you to build and host your own tool layer. This package gives you 20 production-ready MCP tools out of the box.

Install

npx @purplesquirrel/nvidia-mcp

Or install globally:

npm install -g @purplesquirrel/nvidia-mcp
nvidia-mcp

Configure

Add to your MCP config (claude_desktop_config.json or ~/.mcp.json):

{
  "mcpServers": {
    "nvidia": {
      "command": "npx",
      "args": ["-y", "@purplesquirrel/nvidia-mcp"],
      "env": {
        "NVIDIA_API_KEY": "nvapi-..."
      }
    }
  }
}

Get a free API key at build.nvidia.com -- no credit card required, ~40 requests/minute.

Tools (20)

| Tool | Description | Key Params | |------|-------------|------------| | nvidia_codereview | AI code review for security and quality using Qwen 2.5 Coder 32B | code, language | | nvidia_compare | Semantic text comparison using Nemotron embeddings -- similarity scores and candidate ranking | text_a, text_b or query + candidates | | nvidia_detect | Object detection in images using Grounding DINO with Llama 3.2 90B Vision fallback | image_path, prompt | | nvidia_finance | Financial and market analysis using Nemotron Ultra 253B | query, context | | nvidia_guardrails | Enhanced guardrails using 6 safety models with weighted scoring and topic control | text, direction | | nvidia_ladder | Benchmark a prompt across the full Nemotron family (Nano 9B to Ultra 253B) | prompt | | nvidia_multimodal | Cross-modal semantic search using NVCLIP + Nemotron embeddings | query, documents | | nvidia_ocr | Extract text from images using NVIDIA OCDRNet with VLM fallback | image_path | | nvidia_parse | Parse documents into structured sections, tables, and metadata | content or image_path | | nvidia_pii | Detect and redact PII (names, emails, SSNs, phone numbers) from text | text | | nvidia_podcast | Convert text into a two-person podcast script (HOST + EXPERT) | topic, content | | nvidia_quality | Quality control pipeline across 3 models (Nano 9B, Super 49B, Super 120B) | prompt | | nvidia_rag | Retrieval-Augmented Generation with in-memory vector store | query, passages | | nvidia_research | Multi-model deep research across Ultra 253B, Super 120B, and Super 49B | topic | | nvidia_reward | Score LLM response quality using Nemotron 70B Reward model | prompt, response | | nvidia_router | Route a prompt to the optimal NVIDIA model by task type | prompt, execute | | nvidia_safety | Content moderation using 4 NVIDIA safety models in sequence | text | | nvidia_scan | Batch security scanner for up to 20 code files | files | | nvidia_translate | Translate text using NVIDIA Riva Translate 4B | text, target_language | | nvidia_video | Video analysis -- frame extraction, scene description via Llama 3.2 90B Vision, timeline summary | video_path |

Why This One?

• Free tier, no catch. NVIDIA NIM API gives ~40 RPM across 189 models at zero cost. No credit card, no usage caps for individual tools.
• Only standalone NIM MCP. No other package wraps NVIDIA's AI pipelines as MCP tools. The alternative is building your own integration from scratch using NeMo or raw API calls.
• 20 tools, one install. Code review, security scanning, RAG, content safety, PII detection, translation, OCR, video analysis, financial analysis, and model benchmarking -- all from a single npx command.

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | NVIDIA_API_KEY | Yes | NIM API key from build.nvidia.com |

Development

git clone https://github.com/ExpertVagabond/nvidia-mcp.git
cd nvidia-mcp
npm install
npm run build
npm start

License

MIT -- Purple Squirrel Media