Vision Mcp Server

Vision MCP Server | 图片分析 MCP

[English](#english) | [中文](#中文)

---

中文

一个用于图片分析的 MCP (Model Context Protocol) 服务器，支持图片内容分析和描述。 例如当你在客户端的模型只支持文字输入，这时你可以使用视觉模型mcp来弥补。 这个项目采用了魔搭社区免费的视觉模型Qwen3-VL-30B-A3B-Instruct（你也可以在配置中，使用魔搭社区自行更换为自己想要的视觉模型）。

功能特点

• 支持本地图片文件和在线图片 URL
• 基于魔搭社区 AI 模型的智能图像分析
• 完全兼容 MCP 协议
• TypeScript 支持，提供完整的类型定义

安装

方式一：使用 npx（推荐）

无需预先安装，在客户端填写以下内容npx 会自动下载并运行最新版本：

{
  "mcpServers": {
    "vision-mcp-server": {
      "command": "npx",
      "args": [
        "-y",
        "vision-mcp-server"
      ],
      "env": {
        "MODELSCOPE_TOKEN": "your_modelscope_token_here",
        "MODELSCOPE_MODEL": "Qwen/Qwen3-VL-30B-A3B-Instruct"
      }
    }
  }
}

方式二：全局安装

npm install -g vision-mcp-server

然后在客户端配置中：

{
  "mcpServers": {
    "vision-mcp-server": {
      "command": "vision-mcp-server",
      "env": {
        "MODELSCOPE_TOKEN": "your_modelscope_token_here",
        "MODELSCOPE_MODEL": "Qwen/Qwen3-VL-30B-A3B-Instruct"
      }
    }
  }
}

方式三：本地安装

npm install vision-mcp-server

然后在客户端配置中：

{
  "mcpServers": {
    "vision-mcp-server": {
      "command": "node",
      "args": ["node_modules/vision-mcp-server/dist/index.js"],
      "env": {
        "MODELSCOPE_TOKEN": "your_modelscope_token_here",
        "MODELSCOPE_MODEL": "Qwen/Qwen3-VL-30B-A3B-Instruct"
      }
    }
  }
}

环境变量配置

在使用前，需要设置以下环境变量：

• MODELSCOPE_TOKEN: 魔搭社区的 API 密钥（必需）
• 获取方式：访问 魔搭社区 → 个人中心 → API令牌
• MODELSCOPE_MODEL: 使用的模型名称（可选，默认为 "Qwen/Qwen3-VL-30B-A3B-Instruct"）
• 支持其他视觉模型，如：Qwen/Qwen2-VL-7B-Instruct

使用示例

// 分析本地图片
{
  "name": "analyze_image",
  "arguments": {
    "image": "/path/to/your/image.jpg",
    "prompt": "请描述这张图片的内容"
  }
}

// 分析在线图片
{
  "name": "analyze_image",
  "arguments": {
    "image": "https://example.com/image.jpg",
    "prompt": "这张图片中有哪些物体？"
  }
}

API 参考

analyze_image

分析图片内容并提供详细描述。

参数:

• image (string): 图片 URL 或本地文件路径
• prompt (string, 可选): 对图片的问题或分析要求，默认为 "请描述这张图片的内容"

返回: 图片内容的详细文本描述。

开发

构建

npm run build

测试

npm test

贡献

欢迎提交 Issue 和 Pull Request！

许可证

[MIT](LICENSE)

更新日志

1.0.0

• 初始版本发布
• 支持图片分析功能
• 兼容 MCP 协议

---

English

A Vision Analysis MCP (Model Context Protocol) Server that supports image content analysis and description.

Features

• Support for local image files and online image URLs
• Intelligent image analysis based on ModelScope AI models
• Full compatibility with MCP protocol
• TypeScript support with complete type definitions

Installation

Option 1: Using npx (Recommended)

No need to pre-install, npx will automatically download and run the latest version:

{
  "mcpServers": {
    "vision-mcp-server": {
      "command": "npx",
      "args": [
        "-y",
        "vision-mcp-server"
      ],
      "env": {
        "MODELSCOPE_TOKEN": "your_modelscope_token_here",
        "MODELSCOPE_MODEL": "Qwen/Qwen3-VL-30B-A3B-Instruct"
      }
    }
  }
}

Option 2: Global Installation

npm install -g vision-mcp-server

Then in your client configuration:

{
  "mcpServers": {
    "vision-mcp-server": {
      "command": "vision-mcp-server",
      "env": {
        "MODELSCOPE_TOKEN": "your_modelscope_token_here",
        "MODELSCOPE_MODEL": "Qwen/Qwen3-VL-30B-A3B-Instruct"
      }
    }
  }
}

Option 3: Local Installation

npm install vision-mcp-server

Then in your client configuration:

{
  "mcpServers": {
    "vision-mcp-server": {
      "command": "node",
      "args": ["node_modules/vision-mcp-server/dist/index.js"],
      "env": {
        "MODELSCOPE_TOKEN": "your_modelscope_token_here",
        "MODELSCOPE_MODEL": "Qwen/Qwen3-VL-30B-A3B-Instruct"
      }
    }
  }
}

Environment Variables Configuration

Before using, you need to set the following environment variables:

• MODELSCOPE_TOKEN: ModelScope API key (required)
• Get it from: ModelScope → Profile → API Token
• MODELSCOPE_MODEL: Model name to use (optional, default is "Qwen/Qwen3-VL-30B-A3B-Instruct")
• Supports other vision models, such as: Qwen/Qwen2-VL-7B-Instruct

Usage Examples

// Analyze local image
{
  "name": "analyze_image",
  "arguments": {
    "image": "/path/to/your/image.jpg",
    "prompt": "Please describe the content of this image"
  }
}

// Analyze online image
{
  "name": "analyze_image",
  "arguments": {
    "image": "https://example.com/image.jpg",
    "prompt": "What objects are in this image?"
  }
}

API Reference

analyze_image

Analyze image content and provide detailed description.

Parameters:

• image (string): Image URL or local file path
• prompt (string, optional): Question or analysis requirement for the image, default is "Please describe the content of this image"

Returns: Detailed text description of the image content.

Development

Build

npm run build

Test

npm test

Contributing

Issues and Pull Requests are welcome!

License

[MIT](LICENSE)

Changelog

1.0.0

• Initial release
• Image analysis support
• MCP protocol compatibility

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Markusbetter
• Source: Markusbetter/vision-mcp-server
• License: MIT

Install and usage instructions live in the source repository linked above.