NIM MCP

A Model Context Protocol (MCP) server that wraps the NVIDIA NIM (NVIDIA Inference Microservices) API (build.nvidia.com) to provide chat completions and model discovery, featuring an interactive terminal TUI for API key profile and default model management.

Features

• chat_completion Tool: Generate chat completions using any model hosted on the NVIDIA NIM catalog. Supports standard text models, vision/VLM input (e.g., passing base64 images), custom temperatures, token limits, and function/tool calling.
• list_models Tool: Automatically fetch and list all available models through the active API key.
• Dynamic Config Reloading: The MCP server dynamically reloads config.json on every call. Switching profiles or default models via the TUI takes effect immediately, without requiring a server restart.
• Single API Key, Full Access: A single NVIDIA NIM API key covers the entire catalog of models. You can easily switch models by simply passing a different model string (e.g., z-ai/glm-5.2, meta/llama-3.1-405b-instruct, qwen/qwen3-coder-480b-a35b-instruct).
• Interactive Terminal TUI: A beautiful React Ink terminal interface to:
  • Create, edit, and delete multiple API key profiles.
  • View, search, and navigate available NIM models grouped by vendor.
  • Set a default model (marked with ★) to be resolved automatically when none is specified in client requests.
• Self-Contained Windows Setup: Includes automated scripts to build a local environment (Python virtual environment, portable Node.js, package installs) and create a desktop shortcut.

Prerequisites

• OS: Windows (setup scripts and shortcut generation are Windows-based)
• Python: version 3.10 or higher
• NVIDIA NIM API Key: A free API key from build.nvidia.com

Quick Start

1. Setup the Environment

Double-click or execute the setup script in your terminal:

setup.bat

This will automatically:

1. Create a default config.json configuration file.
2. Verify Python is installed and configure a local Python virtual environment (.venv).
3. Download a portable Node.js runtime (.node_venv) and install Ink dependencies for the TUI.
4. Generate a desktop shortcut called NIM MCP.

2. Configure Profiles via the TUI

Run run.bat (or open the NIM MCP desktop shortcut) to start the TUI.

1. Configure an API Key:
   • In the API KEY tab, navigate the left panel actions to select Create API key.
   • Type a profile name (e.g., Default) and press Enter.
   • Paste your NVIDIA NIM API key (nvapi-...) and press Enter.
   • Go to Save using the arrow keys and press Enter.
   • Select the newly created profile and choose Switch to selected to make it active.
2. Select a Default Model:
   • Navigate to the MODELS tab (using Tab or arrow keys). The TUI will fetch the live catalog from the NVIDIA endpoint.
   • Use the arrow keys to browse through vendors (left panel) and their models (right panel).
   • Press Enter on any model (such as z-ai/glm-5.2) to set it as your default model.

3. Register the MCP Server

Add the server configuration to your claude_desktop_config.json (typically located at %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "nim": {
      "command": "<path-to-NIM_MCP>\\.venv\\Scripts\\python.exe",
      "args": [
        "<path-to-NIM_MCP>\\server.py"
      ]
    }
  }
}

Note: You do not need to specify NIM_API_KEY in the environment variables here. The server will dynamically read the active profile and API key from config.json at runtime.

TUI Preview

License

This project is licensed under the MIT License.