ddddocr

Go to https://github.com/86maid/ddddocr/releases and download the prebuilt binary for your operating system. On Windows this will be ddddocr.exe; on Linux/macOS it is a plain executable. Make it executable if needed.

chmod +x ./ddddocr

Launch the server with the --mcp flag alongside the capabilities you want to expose. The --ocr flag enables text recognition, --det enables object detection, and --slide enables slider-captcha matching. The server will listen on the specified address.

./ddddocr --address 0.0.0.0:8000 --ocr --det --slide --mcp

Send an initialize request to the MCP endpoint to confirm the server is responding. The server supports MCP protocol version 2025-11-25.

curl -X POST http://localhost:8000/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":0,"method":"initialize","params":{}}'

Add the ddddocr server to your MCP client configuration. Since the server runs as an HTTP service, you connect to it via its HTTP endpoint rather than spawning it as a subprocess.

{
  "mcpServers": {
    "ddddocr": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

Call the ocr tool with a base64-encoded image. The optional color_filter parameter narrows recognition to specific colored text (e.g., 'green', 'red'), which helps with colored captchas.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "ocr",
    "arguments": {
      "image": "<base64_encoded_image_string>",
      "color_filter": "green"
    }
  }
}