visionMCP
visionMCP 👁️
更大推理LLM的眼睛。
visionMCP 是一个 Model Context Protocol 服务器,为任何支持 MCP 的智能体提供真正的视觉能力。纯文本推理模型可以将任何它无法看到的内容委托给此服务器:描述截图、回答关于照片的问题、OCR 文档或比较两张图片——服务器负责“看”,然后返回文本。
它支持所有三种主流视觉后端,在运行时通过单个 config.json 选择:
提供商 | API | 示例模型 |
Ollama | OpenAI 兼容 ( |
|
OpenAI | Chat Completions 视觉 API |
|
Anthropic | Claude Messages 视觉 API |
|
功能特性
🔍 四个视觉工具供推理 LLM 调用:
describe_image— 完整的自然语言描述ask_about_image— 针对任意图片的定向问答extract_text— OCR / 转录compare_images— 并排比较
🖼️ 接受所有来源:本地文件路径、
http(s)URL 以及 base64data:URI。📦 零图片预处理:超大图片会自动缩小并重新编码为 JPEG,以适配提供商的有效载荷限制。
🔌 三种传输方式:
stdio(默认,用于本地 MCP 客户端)、http(可流式 HTTP,用于远程托管)或sse(传统 Server-Sent Events)。⚙️ 一个
config.json控制提供商、API 密钥、API URL 和模型。环境变量和 CLI 标志可以覆盖任何设置。🚀 由
uv管理,可安装、可运行、可托管。
Related MCP server: depu-img-mcp
快速开始
1. 安装
需要 uv 和 Python ≥ 3.10。
cd visionMCP
uv sync2. 配置
附带的 config.json 已可直接用于本地 Ollama。通过编辑文件切换提供商:
// config.json
{
"provider": "openai", // "ollama" | "openai" | "anthropic"
"api_key": "sk-...", // or leave "" and export OPENAI_API_KEY
"api_url": "", // "" = provider default
"model": "" // "" = provider default
}查看 docs/configuration.md 了解所有选项,以及 docs/providers.md 了解每个提供商的设置。
安全: 不要将真实的 API 密钥提交到 git——将
config.json复制为config.local.json(自动忽略)或使用环境变量。服务器永远不会记录你的密钥。
3. 运行
uv run vision-mcp # stdio transport (default)
uv run vision-mcp --transport http --host 0.0.0.0 --port 8100 # host remotely
uv run vision-mcp --show-config # print resolved config (key masked)接入 MCP 客户端
opencode (opencode.json)
{
"mcpServers": {
"visionMCP": {
"type": "stdio",
"command": "uv",
"args": ["run", "--directory", "/absolute/path/to/visionMCP", "vision-mcp"]
}
}
}Claude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"visionMCP": {
"command": "uv",
"args": ["run", "--directory", "/absolute/path/to/visionMCP", "vision-mcp"]
}
}
}通用 MCP 客户端 (stdio)
{
"mcpServers": {
"visionMCP": {
"command": "/path/to/visionMCP/.venv/bin/vision-mcp",
"args": ["--config", "/path/to/visionMCP/config.json"]
}
}
}服务器除了工具调用提供的内容外,不会向视觉 API 发送图片内容。图片字节保留在内存中,不会写入磁盘。
工具参考
工具 | 参数 | 返回 |
|
| 纯文本的完整描述 |
|
| 针对性回答 |
|
| 转录/OCR 后的文本 |
|
| 纯文本的比较结果 |
| — | 提供商、模型、传输方式 |
image 参数接受以下任意一种:
/path/to/photo.png # local file
https://example.com/x.jpg # URL (downloaded at call time)
data:image/png;base64,iVBORw0KGgo... # base64 data URI工作原理
所有功能都汇聚在 src/vision_mcp/pipeline.py 中的一个函数中:
image (path / URL / data URI) → base64 + mime → vision model → text
_read() _encode() API[provider] look()服务器工具是薄封装层:pipeline.look(cfg, [image], prompt)。
文档
开发
uv sync --group dev
uv run ruff check .
uv run pytest许可证
MIT
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseAqualityCmaintenanceGive MCP-compatible AI agents image analysis, metadata inspection, cropping, OCR, and image comparison through any OpenAI-compatible vision model.6MIT
- AlicenseAqualityBmaintenanceEnables text-only language models to understand images by forwarding image and prompt requests to vision model backends via MCP, returning descriptive text. Supports multiple OpenAI-compatible providers with task routing and safety checks.1MIT
- AlicenseAqualityBmaintenanceEnables non-multimodal models to see images by providing MCP tools for image understanding and OCR, backed by any OpenAI-compatible vision model.2MIT
- AlicenseAqualityBmaintenanceEnables any MCP client to perform image understanding and OCR via any OpenAI-compatible vision-language model. Supports local, private inference without images leaving the machine.256MIT
Related MCP Connectors
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Generate on-brand images from your AI agent: design, edit, and render templates over MCP.
Social media analytics, video analysis, and competitor intel for any MCP-compatible AI agent.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/ThisUserIsRandom/visionMCP'
If you have feedback or need assistance with the MCP directory API, please join our Discord server