deepsee
MCP server that gives text-only LLMs like DeepSeek vision capabilities by converting images to text via vision APIs, enabling image description, OCR, and generation in MCP clients.
README
deepsee
给 DeepSeek 等纯文本模型补上「看图」能力 —— 在 Claude Code / Codex / Cursor / 任意聊天客户端里贴一张图,模型能读懂并回答,不报 unsupported,和原生识图模型一样丝滑。
原理:DeepSeek 纯文本看不到图,deepsee 用便宜/免费的视觉 API(默认智谱 GLM-4V-Flash)把图转成文字描述,再喂给你的 DeepSeek。视觉供应商和上游 LLM 都任意自接,不绑定任何厂商。
你贴图 ──→ 视觉 API(免费/便宜)转成文字 ──→ 你的 DeepSeek 读描述回答
快速开始
pip install deepsee-mcp # 或: git clone <repo> && ./install.sh
deepsee setup # 向导:①选视觉供应商(智谱免费) ②填上游 ③集成
deepsee install # 装进 Claude Code(Hook + MCP),重启即生效
# 然后:Claude Code 里直接贴图,DeepSeek 就能看懂
完整的配置方法(拿 key、各字段含义、常见组合、排障)见 docs/CONFIGURATION.md。
没有 API key?
deepsee setup 选「演示模式」或「本地 OCR」即可零成本先跑起来,之后随时补 key。
六种使用形态
| 形态 | 命令/方式 | 适用 |
|---|---|---|
| Claude Code 路由代理(主) | deepsee route on |
覆盖所有图片场景:粘贴/headless/Read,协议层转文字 |
| Claude Code Hook | deepsee install |
交互式贴图/Read 自动识图(写在 settings.local.json,cc-switch 不抹) |
| MCP 工具 | deepsee install → describe_image/generate_image |
只靠视觉 key 即可,不依赖对话额度 |
| 网页聊天界面 | deepsee ui / deepsee web |
浏览器里贴图即聊(流式) |
| 双协议代理 | deepsee start |
Cherry Studio / ChatBox / Cursor / OpenCode / Cline |
| Codex skill | deepsee install --skill |
Codex Desktop / CLI |
| CLI | deepsee see/ocr/gen |
终端直接看图/画图 |
装成 MCP 插件(Cursor / Cline / 任意 MCP 客户端)
仓库里自带 mcp.json,在 Cursor(Cursor 设置 → MCP)或 Cline 里指向它即可;
或手动注册(把 deepsee 换成 deepsee mcp 的实际命令):
{
"mcpServers": {
"deepsee": {
"command": "deepsee",
"args": ["mcp"]
}
}
}
- 全局安装后即可用:
pipx install deepsee-mcp(或pip install deepsee-mcp),然后deepsee mcp。 - 工具:
describe_image/ocr_image/generate_image/vision_status。 - 配置目录通过环境变量
DEEPSEE_HOME覆盖(默认~/.deepsee)。 - 发布到 Smithery/mcp.so:见
docs/CONFIGURATION.md的「MCP 发布」一节。
兼容性
- 视觉供应商:任意 OpenAI 兼容视觉端点 —— 智谱 GLM-4V-Flash(免费)、阿里百炼 qwen-vl、硅基流动、OpenAI、Gemini、Groq、OpenRouter、本地 vLLM/Ollama;多供应商自动降级链。
- 上游:任意 OpenAI / Anthropic 兼容端点 —— DeepSeek 官方、各类中转/网关、本地 Ollama;
format=auto时双协议各自流式转发。 - 免 key:
demo演示模式(零网络)、local-ocr本地 OCR(离线)。 - 纯 stdlib,零第三方依赖;Linux / macOS / Windows。
项目结构
deepsee/
├── vision.py 视觉客户端 + 多供应商降级链
├── caption.py 识图(describe/ocr/qa)
├── chat.py 对话服务(图片→文字→上游,代理与网页共用)
├── proxy.py OpenAI + Anthropic 双协议代理(图片块→文字→DeepSeek)
├── mcp_server.py 手写轻量 MCP(零依赖)
├── hook_claude.py Claude Code PreToolUse Hook
├── web.py + web/static/ 网页聊天界面(标签页:对话/识图/画图/状态/配置)
├── cli.py / installer.py / config.py / util.py / generate.py
docs/ PRD / 架构 / 配置方法
tests/ pytest(75 用例,全离线 mock)
开发
pip install -e '.[dev]'
pytest
License
MIT License —— 随便用、随便改,欢迎提交 PR 与 issue。
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.