Skip to main content
Glama

depu-img-mcp

图像理解 MCP 服务器 —— 让不支持图像的文本模型通过 MCP 把图片 + prompt 转发给视觉模型,返回描述文本。自身不做推理,是协议适配与转发层。

特性

  • MCP v2(MCPServer + Streamable HTTP,2026-07-28 规范),同时支持 stdio / SSE

  • Docker 部署(多阶段构建 + docker-compose)

  • 多后端 provider:任意 OpenAI 兼容端点(LiteLLM 网关 / vLLM / OpenAI / DashScope / 智谱 / OpenRouter / Ollama …)

  • 双层 prompt 注入:全局 base_vision_prompt + per-call prompt + task_type 路由(auto/general/ocr/ui/debug/describe)

  • 安全:magic byte 校验、大小限制、SSRF 防护、路径白名单

  • 轻量 Web 后台(/admin):查看配置 + 在线测试图片

  • 配置双源:环境变量覆盖 config.toml,密钥支持 ${ENV} 插值

Related MCP server: Vision MCP

快速开始

# 1. 准备 .env(至少填 ADMIN_TOKEN 和视觉后端 API key)
cp .env.example .env
# 编辑 .env 填入 API key

# 2. Docker 启动(首次会自动生成 ./conf/config.toml 默认配置)
docker compose up -d

# 3. 访问后台改配置(或直接编辑 ./conf/config.toml 后重启)
open http://localhost:8080/admin
# MCP 端点:http://localhost:8080/mcp

首次启动无需预先准备 config.toml:容器挂载 ./conf/ 目录,若里面没有配置文件,会自动从内置模板生成一份。之后可在 /admin 后台在线编辑,或直接改 ./conf/config.toml 后 docker compose restart。

本地运行(开发)

uv sync
uv run python -m depu_img_mcp          # 默认 streamable-http
MCP_TRANSPORT=stdio uv run python -m depu_img_mcp   # stdio 模式给 Claude Desktop 等

MCP 工具

image_understand

image_understand(
    image: str,            # http(s) URL 或 data:image/...;base64,... URI(内联 base64)。不支持本地文件路径(Docker 部署,容器看不到客户端文件系统)
    prompt: str = "",      # 用户问题;空则给出通用描述(物体、文字/OCR、布局、颜色等)
) -> str

provider / model 由后台 admin 统一配置,客户端不能指定。这用于企业内网服务,后端路由由运维通过 /admin 后台决定。

只有一个工具。客户端传图 + prompt,服务器用后台配置的默认 provider 及其 model 进行视觉理解,返回文本描述。

配置示例

[[providers]]
name = "default"
type = "openai-compat"
base_url = "https://gateway.ai.depu.school/v1"
api_key = "${DEPU_GATEWAY_API_KEY}"
model = "Kimi-K2.7-Code"
auth_header = "bearer"

客户端配置(Claude Desktop 示例)

stdio 模式:

{
  "mcpServers": {
    "depu-img": {
      "command": "python",
      "args": ["-m", "depu_img_mcp"],
      "env": { "MCP_TRANSPORT": "stdio", "DEPU_GATEWAY_API_KEY": "sk-..." }
    }
  }
}

HTTP 模式(支持远程 MCP 的客户端):

{
  "mcpServers": {
    "depu-img": { "url": "http://localhost:8080/mcp" }
  }
}

License

MIT

Available Tools

1 tool
image_understandA

Describe or analyze an image using a vision model.

Call this whenever you need to understand an image you cannot see. The backend provider and model are determined by the server admin configuration and cannot be overridden from the client.

Args: image: The image as an http(s) URL, or a data:image/...;base64,... URI (inline base64). Local file paths are NOT supported because the server runs in a container and cannot see the client filesystem — encode the file as a base64 data URI first. prompt: What you want to know about the image. Empty = a general description of the image (objects, text/OCR, layout, colors).

Returns: The vision model's text description of the image.

Inputs: image (required), prompt. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.9/5.0

    Scored across 1 tool

    Disambiguation5/5

    Only one tool exists, so there is no possibility of confusion or overlapping purposes.

    Naming Consistency5/5

    A single tool name is trivially consistent; there are no conflicting naming conventions to assess.

    Tool Count3/5

    One tool for image understanding is thin but acceptable for a single-purpose server; it sits at the low end of the typical range.

    Completeness5/5

    The tool provides comprehensive image understanding (description, OCR, layout, colors) and accepts both URLs and base64, covering the full domain without obvious gaps.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers