eyes-for-deepseek

# Tool List > DeepSeek has no eyes? Give it some. > > An MCP server that unifies access to mainstream free/low-cost visual APIs in China, enabling pure-text LLMs like DeepSeek to gain capabilities such as **image viewing, OCR, and screenshot Q&A**. --- ### One-sentence Install Send this to your AI coding assistant (Claude Code / Cursor / Cline), and it will install everything for you: > **"Install this MCP for me: https://github.com/Luuh2/eyes-for-deepseek , use the Bailian backend"** Replace `Bailian` with `Doubao` or `Zhipu` if you prefer. You'll need an API key for your chosen platform — the AI will guide you through getting one. --- [English](#english) | --- <a name="english"></a> ## English ### One-sentence Install Send this to your AI coding assistant (Claude Code / Cursor / Cline), and it will install everything for you: > **"Install this MCP for me: https://github.com/Luuh2/eyes-for-deepseek , use the Bailian backend"** Replace `Bailian` with `Doubao` or `Zhipu` if you prefer. You'll need an API key for your chosen platform — the AI will guide you through getting one. ### What is this? DeepSeek and many other LLMs can't see images. This MCP server acts as a bridge, routing images to a vision model and returning text descriptions. Your text-only LLM can then "see" through four tools: | Tool | What it does | |------|-------------| | `describe_image` | Detailed description of any image | | `ocr_image` | Extract text from screenshots, documents, tables | | `ask_image` | Ask questions about an image (debug code, analyze charts) | | `summarize_image` | Quick one-sentence summary | ### Supported Vision Backends | Backend | Vision Model | Free Tier | Key Link | |---------|-------------|-----------|----------| | `bailian` (default) | Alibaba Qwen-VL | Yes | [Get Key](https://bailian.console.aliyun.com/) | | `doubao` | ByteDance Doubao Vision | Yes | [Get Key](https://console.volcengine.com/ark/region:ark+cn-beijing/apiKey) | | `zhipu` | Zhipu GLM-4V | Yes | [Get Key](https://open.bigmodel.cn/usercenter/apikeys) | | `custom` | Any OpenAI-compatible API | — | Your own endpoint | All three built-in backends offer **free quotas** for new users. Pick whichever you prefer — they all speak the same OpenAI-compatible protocol. ### Quick Start ```bash # 1. Clone git clone https://github.com/Luuh2/eyes-for-deepseek.git cd eyes-for-deepseek # 2. Install pip install -r requirements.txt # 3. Configure cp .env.example .env # Edit .env: # - Pick a backend: VISION_BACKEND=bailian (or doubao, zhipu) # - Fill in your API key for that backend ``` ### Connect to Claude Code Add to your project's `.mcp.json`: ```json { "mcpServers": { "eyes-for-deepseek": { "command": "python", "args": ["path/to/eyes-for-deepseek/mcp_server.py"] } } } ``` Works with any MCP-compatible client: Cursor, Cline, Continue, etc. ### How it works ``` You: "What's in this screenshot?" DeepSeek (no vision) MCP protocol Eyes for DeepSeek Bailian / Doubao / Zhipu Vision API "It's a login form with..." DeepSeek replies with the description ``` ### Security - Your API key stays in `.env` — it's gitignored and never committed - The server reads `.env` locally at startup - Keys are only sent to your chosen vision API endpoint - No telemetry, no phoning home, no third-party logging ### License MIT