Qwen Vision MCP Server
by kira4094
README.md
# Qwen Vision MCP Server
MCP server for Alibaba Qwen3.7-plus (multimodal vision model), connected via DashScope's **Anthropic-compatible** endpoint.
## Features
- Single tool: `qwen_vision_understand` — analyze an image with Qwen3.7-plus
- Supports local image files (png/jpg/jpeg/gif/webp/bmp) and remote URLs
- Same Anthropic Messages API format as Claude
- Token usage reported back in each response
## Requirements
- Node.js >= 18
- A DashScope API key — get one at https://dashscope.console.aliyun.com/apiKey
## Install
```bash
cd qwen-vision-mcp-server
npm install
```
## Configuration
Environment variables:
| Variable | Required | Default | Description |
|---|---|---|---|
| `DASHSCOPE_API_KEY` | ✅ | — | Your DashScope API key |
| `QWEN_MODEL` | | `qwen3.7-plus` | Model name (e.g. `qwen-vl-plus`, `qwen3-vl-plus`) |
| `QWEN_BASE_URL` | | `https://dashscope.aliyuncs.com/apps/anthropic/v1/messages` | Override endpoint |
| `QWEN_MAX_TOKENS` | | `4096` | Default max output tokens |
## Run locally
```bash
DASHSCOPE_API_KEY=sk-xxx npm start
```
## CC-Switch / Claude Code config
```json
{
"mcpServers": {
"qwen-vision": {
"type": "stdio",
"command": "cmd",
"args": ["/c", "npx", "-y", "qwen-vision-mcp-server"],
"env": {
"DASHSCOPE_API_KEY": "sk-your-dashscope-key",
"QWEN_MODEL": "qwen3.7-plus"
}
}
}
}
```
Or if installed locally:
```json
{
"mcpServers": {
"qwen-vision": {
"type": "stdio",
"command": "node",
"args": ["E:/Projects/Claude/MCP/qwen/qwen-vision-mcp-server/src/index.js"],
"env": {
"DASHSCOPE_API_KEY": "sk-your-dashscope-key"
}
}
}
}
```
## Tool: `qwen_vision_understand`
| Parameter | Type | Required | Description |
|---|---|---|---|
| `image` | string | ✅ | Local file path or HTTP(S) URL |
| `prompt` | string | ✅ | What to ask about the image |
| `max_tokens` | number | | Max output tokens (default 4096) |
| `temperature` | number | | 0-2, default 1 |
Example:
```
Use qwen_vision_understand to analyze this UI mockup:
/Users/me/Downloads/login-page.png
"Recreate this as HTML with Tailwind CSS"
```
## License
MIT
TDQS
A3.6/5.0
Scored across 1 tool
Disambiguation5/5
Only one tool exists, so there is no possibility of confusion or overlap.
Naming Consistency5/5
The single tool follows a clear verb_noun pattern (qwen_vision_understand), consistent with common MCP naming conventions.
Tool Count2/5
A vision MCP server typically requires multiple specialized tools (e.g., describe, analyze, compare) rather than a single monolithic tool. One tool feels under-scoped for the domain.
Completeness3/5
The single tool covers a broad range of vision tasks (UI analysis, charts, documents), but lacks structured decomposition into separate operations, which limits granularity and may cause agent confusion.
Maintenance
ActivityInactive
ResponsivenessNo issues