Skip to main content
Glama
README.md
# image-mcp

基于 DashScope 视觉模型的本地 MCP server,用于识别本地或远程图片,并默认以 XML 形式返回图片内容,帮助 AI 理解图片。

## 安装

```sh
npm install
```

## MCP 配置

模型配置通过 MCP client 的 `env` 注入到 server 进程,不需要也不应该作为 tool 参数传递。

```json
{
  "mcpServers": {
    "image-mcp": {
      "command": "node",
      "args": ["/absolute/path/to/image-mcp/src/index.mjs"],
      "env": {
        "DASHSCOPE_API_KEY": "your-api-key",
        "VISION_MODEL": "qwen-vl-plus",
        "DASHSCOPE_BASE_URL": "https://dashscope.aliyuncs.com/compatible-mode/v1"
      }
    }
  }
}
```

`DASHSCOPE_BASE_URL` 可省略,默认使用 `https://dashscope.aliyuncs.com/compatible-mode/v1`。

## Tool

`recognize_image`

- `image_path`: 本地图片路径,和 `image_url` 二选一。
- `image_url`: 远程图片 URL,和 `image_path` 二选一。
- `prompt`: 可选识图提示词,默认值为 `识别图像内容,使用xml形式输出,帮助ai理解图片内容`。
- `max_tokens`: 可选最大输出 token 数,默认 `1024`。

## CLI

```sh
npm run vision -- ./dist/ScreenShot_2026-08-01_155219_072.png "识别图像内容,使用xml形式输出,帮助ai理解图片内容"
npm run vision -- --url https://example.com/image.png
```

## 本地 npm 使用

如果只是本机使用,不需要发布到 npm registry。推荐用 `npm link` 或 `npm pack`。

### 方式一:npm link

在项目目录执行:

```sh
npm install
npm link
```

然后 MCP 配置可以直接使用全局命令:

```json
{
  "mcpServers": {
    "image-mcp": {
      "command": "image-mcp",
      "args": [],
      "env": {
        "DASHSCOPE_API_KEY": "your-api-key",
        "VISION_MODEL": "qwen-vl-plus",
        "DASHSCOPE_BASE_URL": "https://dashscope.aliyuncs.com/compatible-mode/v1"
      }
    }
  }
}
```

如果之后修改了源码,一般不需要重新 link;重启 MCP client 即可加载新代码。

取消本地链接:

```sh
npm unlink -g image-mcp
```

### 方式二:npm pack

如果想模拟正式 npm 包安装,但仍然只在本地使用:

```sh
npm install
npm pack
npm install -g ./image-mcp-1.0.0.tgz
```

安装后同样可以在 MCP 配置中使用:

```json
{
  "mcpServers": {
    "image-mcp": {
      "command": "image-mcp",
      "args": [],
      "env": {
        "DASHSCOPE_API_KEY": "your-api-key",
        "VISION_MODEL": "qwen-vl-plus"
      }
    }
  }
}
```

升级本地包时重新执行:

```sh
npm pack
npm install -g ./image-mcp-1.0.0.tgz
```

### 检查命令

确认全局命令可用:

```sh
which image-mcp
image-mcp
```

`image-mcp` 是 MCP stdio server,直接运行后会等待 MCP client 输入;没有输出不代表失败。实际验证建议在 MCP client 中查看是否能列出 `recognize_image` 工具。


## 特殊说明

项目扩展自 : https://github.com/asuojun/claude-vision-skill/tree/master

TDQS

A4/5.0

Scored across 1 tool

Disambiguation5/5

Only one tool exists, so there is no possibility of confusion between tools. The single recognize_image tool has a clear and distinct purpose.

Naming Consistency5/5

The tool name recognize_image follows a clear verb_noun pattern, which is descriptive and predictable. With only one tool, the naming convention is inherently consistent.

Tool Count3/5

Having a single tool feels thin for an image-related server. While the tool covers the core recognition function, the server lacks any additional utilities, making it borderline in scope.

Completeness5/5

The server's stated purpose is to help AI understand image content, and recognize_image fulfills this completely. There are no obvious missing operations within this narrow domain.

Maintenance

ActivityStale
ResponsivenessNo issues