DINO-X

DINO-X MCP Server provides multimodal computer vision capabilities built on top of the DINO-X models. It supports fine-grained object detection, region-level descriptions, and structured outputs that include object categories, counts, locations, and attributes. The server exposes several tools through the MCP interface: full-scene object detection (detect-all-objects), text-prompted object detection (detect-objects-by-text), human pose estimation (detect-human-pose-keypoints), and a visualization utility (visualize-detection-result) that saves annotated images locally. You can run the MCP locally via STDIO or expose it over HTTP in streamable mode, enabling integration with other MCP servers to form end-to-end visual agents or automation pipelines.

Prerequisites:

• Node.js (LTS version) and npm installed on your machine
• Optional: a valid DINO-X API key if you plan to use the API-key protected features

Option B — Use the NPM package locally (STDIO)

1. Install Node.js from https://nodejs.org/ or ensure npm is available
2. Install and run the MCP via npx (as configured in mcp_config):

# Quick run using npx and the official package
# Ensure you have an API key ready
# Replace variables in the configuration as needed

3. Configure your MCP client with the following example (mcpServers.dinox-mcp):

{
  "mcpServers": {
    "dinox-mcp": {
      "command": "npx",
      "args": ["-y", "@deepdataspace/dinox-mcp"],
      "env": {
        "DINOX_API_KEY": "your-api-key-here",
        "IMAGE_STORAGE_DIRECTORY": "/path/to/your/image/directory"
      }
    }
  }
}

Option C — Run from source locally

1. Ensure Node.js is installed
2. Clone the repository and install dependencies:

git clone https://github.com/IDEA-Research/DINO-X-MCP.git
cd DINO-X-MCP
npm install

3. Build the project:

npm run build

4. Run the server via Node.js and point your MCP client to the built index.js:

{
  "mcpServers": {
    "dinox-mcp": {
      "command": "node",
      "args": ["/path/to/DINO-X-MCP/build/index.js"],
      "env": {
        "DINOX_API_KEY": "your-api-key-here",
        "IMAGE_STORAGE_DIRECTORY": "/path/to/your/image/directory"
      }
    }
  }
}

Note: If you prefer a hosted or containerized setup, you can adapt these steps to a Docker workflow or a cloud deployment, but the readme primarily demonstrates the npm-based local run and the source build workflow.

Tips and caveats:

• The DINO-X MCP supports two transport modes: STDIO (default) and Streamable HTTP. Choose the mode that fits your workflow via command-line flags (e.g., --http for HTTP mode).
• Required environment variable: DINOX_API_KEY is required for API-enabled features. IMAGE_STORAGE_DIRECTORY is optional for STDIO mode and controls where annotated images are saved.
• If you enable HTTP mode behind a gateway, you may also set AUTH_TOKEN to restrict access.
• The npm package to use is @deepdataspace/dinox-mcp. When using npx, the package name is passed as an argument as shown in the examples.
• The client configuration examples assume the MCP server is reachable at localhost:3020 by default when using the built/index.js entry (adjust port as needed).
• If you’re using the Option B or C configurations, ensure proper pathing for /path/to/your/image/directory and /path/to/DINO-X-MCP/build/index.js in your environment.