MCP 服务器

DINO-X Image Detection MCP Server

Empower LLMs with fine-grained visual understanding — detect, localize, and describe anything in images with natural language prompts.

README

DINO-X MCP

English | 中文

Enables large language models to perform fine-grained object detection and image understanding, powered by DINO-X and Grounding DINO 1.6 API.

</p>

💡 Why DINO-X MCP?

Although multimodal models can understand and describe images, they often lack precise localization and high-quality structured outputs for visual content.

With DINO-X MCP, you can:

🧠 Achieve fine-grained image understanding — both full-scene recognition and targeted detection based on natural language.

🎯 Accurately obtain object count, position, and attributes, enabling tasks such as visual question answering.

🧩 Integrate with other MCP Servers to build multi-step visual workflows.

🛠️ Build natural language-driven visual agents for real-world automation scenarios.

🎬 Use Case

🎯 Scenario	📝 Input	✨ Output
Detection & Localization	💬 Prompt:<br>`Detect the fire areas in the forest and visualize with Canvas`<br><br>🖼️ Input Image:<br><img src="/assets/examples/1-1.jpg" width="280" alt="Original forest fire image"/>	<img src="/assets/examples/1-2.png" width="400" alt="Fire detection visualization result"/>
Object Counting	💬 Prompt:<br>`Please analyze this warehouse image, detect all the cardboard boxes, count the total number, and create a complete Canvas visualization webpage.`<br><br>🖼️ Input Image:<br><img src="/assets/examples/2-1.jpeg" width="280" alt="Warehouse image"/>	<img src="/assets/examples/2-2.png" width="400" alt="Box detection result"/>
Feature Detection	💬 Prompt:<br>`Find all red cars in the image and visualize with Canvas`<br><br>🖼️ Input Image:<br><img src="/assets/examples/4-1.jpg" width="280" alt="Cars image"/>	<img src="/assets/examples/4-3.png" width="400" alt="Red car detection result"/>
Attribute Reasoning	💬 Prompt:<br>`Find the tallest person in the image, describe their clothing, and visualize the result with Canvas`<br><br>🖼️ Input Image:<br><img src="/assets/examples/5-1.jpg" width="280" alt="People image"/>	<img src="/assets/examples/5-3.png" width="400" alt="Person detection result"/>
Full Scene Detection	💬 Prompt:<br>`Find the fruit with the highest vitamin C content in the image`<br><br>🖼️ Input Image:<br><img src="/assets/examples/6-1.png" width="280" alt="Fruits image"/>	<img src="/assets/examples/6-3.png" width="400" alt="Fruit detection result"/><br><br>Answer: Kiwi fruit (93mg/100g)
Pose Analysis	💬 Prompt:<br>`Please analyze what yoga pose this is and overlay the keypoints on the original image using canvas`<br><br>🖼️ Input Image:<br><img src="/assets/examples/3-1.jpg" width="280" alt="Yoga pose image"/>	<img src="/assets/examples/3-3.png" width="400" alt="Pose detection result"/>

🚀 Quick Start

1. Prerequisites

Make sure you have Node.js installed. If you don't have Node.js, download it from nodejs.org.

Also, choose an AI assistants and applications that support the MCP Client, including but not limited to:

2. Configure MCP Sever

You can use DINO-X MCP server in two ways:

Option A: Using NPM Package 👍

Add the following configuration in your MCP client:

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

Option B: Using Local Project

First, clone and build the project:

# Clone the project
git clone https://github.com/IDEA-Research/DINO-X-MCP.git
cd DINO-X-MCP

# Install dependencies
pnpm install

# Build the project
pnpm run build

Then configure your MCP client:

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

3. Get API Key

Get your API key from DINO-X Platform (A free quota is available for new users).

Replace your-api-key-here in the configuration above with your actual API key.

4. Available Tools

Restart your MCP client, and you should be able to use the following tools:

Method Name	Description	Input	Output
`detect-all-objects`	Detects and localizes all recognizable objects in an image.	Image	Category names + bounding boxes + captions
`object-detection-by-text`	Detects and localizes objects in an image based on a natural language prompt.	Image + Text prompt	Bounding boxes + object captions
`detect-human-pose-keypoints`	Detects 17 human body keypoints per person in an image for pose estimation.	Image	Keypoint coordinates and captions