vision-mcp-server-pro 1.0.0

MCP server for vision capabilities with automatic model fallback via ModelScope API

vision-mcp-server-pro

MCP server for vision capabilities with automatic model fallback via ModelScope API.

Features

• Analyze images via URL or local file path
• Automatic model fallback: when the primary model hits rate limits, it automatically tries the next model
• Configurable model list via environment variables

Default Model List (fallback order)

1. Qwen/Qwen3.5-397B-A17B
2. Qwen/Qwen3-VL-235B-A22B-Instruct
3. moonshotai/Kimi-K2.5
4. Qwen/Qwen3.5-122B-A10B

Usage

Add to your MCP client (e.g., Claude Code) using uvx (no global install needed):

claude mcp add vision-mcp-server-pro --scope user \
  -e MODELSCOPE_TOKEN=your_token_here \
  -- uvx vision-mcp-server-pro

Environment Variables

使用前，需要设置以下环境变量：

Variable	Required	Description
MODELSCOPE_TOKEN	Yes	魔搭社区的 API 密钥
MODELSCOPE_MODEL	No	首选模型（默认使用回落列表中的第一个）
MODELSCOPE_FALLBACK_MODELS	No	自定义回落模型列表，逗号分隔（覆盖默认列表）

获取 MODELSCOPE_TOKEN

访问 魔搭社区 → 个人中心 → API令牌

Specify Primary Model

claude mcp add vision-mcp-server-pro --scope user \
  -e MODELSCOPE_TOKEN=your_token_here \
  -e MODELSCOPE_MODEL=Qwen/Qwen3-VL-235B-A22B-Instruct \
  -- uvx vision-mcp-server-pro

Custom Fallback Models

claude mcp add vision-mcp-server-pro --scope user \
  -e MODELSCOPE_TOKEN=your_token_here \
  -e MODELSCOPE_FALLBACK_MODELS="Qwen/Qwen3.5-397B-A17B,Qwen/Qwen3-VL-235B-A22B-Instruct" \
  -- uvx vision-mcp-server-pro

Tool

The server provides an analyze_image tool:

• image (required): Image URL or local file path
• prompt (optional): Question or analysis request (default: "请描述这张图片的内容")

License

MIT