glm-vision-mcp
# GLM Vision MCP Server
基于 [智谱 GLM-4.6V-Flash](https://open.bigmodel.cn/) 模型的图像识别 MCP 工具。
## 安装
### 方式一:PyPI 安装
```bash
pip install glm-vision-mcp
```
### 方式二:本地安装
```bash
git clone <repo-url>
cd glm-vision-mcp
pip install -e .
```
## 前置条件
1. 注册 [智谱开放平台](https://open.bigmodel.cn/) 账号
2. 获取 API Key
## 配置
在 MCP Client 配置文件中添加(如 `claude_desktop_config.json` 或 `.mcp.json`):
```json
{
"mcpServers": {
"glm-vision": {
"command": "uv",
"args": ["run", "--directory", "<项目路径>", "glm-vision-mcp"],
"env": {
"ZHIPU_API_KEY": "your-api-key-here"
}
}
}
}
```
或 PyPI 安装后:
```json
{
"mcpServers": {
"glm-vision": {
"command": "python",
"args": ["-m", "glm_vision_mcp.server"],
"env": {
"ZHIPU_API_KEY": "your-api-key-here"
}
}
}
}
```
## 使用
### analyze_image
分析图片内容,支持 URL 和本地文件路径。
**参数:**
| 参数 | 类型 | 必填 | 说明 |
|------|------|------|------|
| `image` | string | 是 | 图片 URL(https://...)或本地文件路径 |
| `prompt` | string | 否 | 自定义分析指令,默认"请详细描述这张图片" |
**示例:**
- 描述图片:`{"image": "https://example.com/photo.jpg"}`
- OCR 提取:`{"image": "C:\\screenshots\\receipt.png", "prompt": "提取图中的所有文字"}`
- 图表分析:`{"image": "/data/chart.png", "prompt": "分析这个图表的趋势和关键数据"}`
TDQS
Scored across 1 tool
With only one tool, there is no possibility of confusion or overlap. The tool's purpose is clearly defined as image analysis, so an agent can unambiguously select it for any image-related task.
The single tool 'analyze_image' follows a clean verb_noun pattern, which is consistent with common API naming conventions. While there are no other tools to compare, the name is descriptive and predictable.
The server has only one tool, which is slightly below the typical 3-15 range, but it is tailored specifically for vision-based image analysis. The single tool is reasonably scoped for this narrow purpose, and adding more tools could dilute its focus.
The tool covers the core and likely only operation needed for image analysis: accepting an image URL or local path and allowing custom instructions. There are no obvious gaps or dead ends for this domain, as the tool provides a complete workflow for its intended use.