PaddleOCR-VL Skill

GPU-accelerated document OCR using the PaddleOCR-VL model running inside an ephemeral Docker container. Auto-detects NVIDIA GPU architecture (Blackwell SM120 vs. standard) and selects the correct official image.

When to Use

• User provides an image path and asks to "read", "OCR", or "extract text" from it
• User wants to parse a document screenshot, newspaper page, or classical text
• User asks about the content of an image file

Architecture

This skill includes an MCP server (server.py) that exposes three tools:

Setup

1. Install the MCP Server

Add to ~/.config/Claude/claude_desktop_config.json (Claude Desktop) or ~/.claude/settings.json (Claude Code):

{
  "mcpServers": {
    "paddle-ocr-vl": {
      "command": "python3",
      "args": ["\x3CINSTALL_DIR>/server.py"]
    }
  }
}

2. Pull the Docker Image (one-time)

# Blackwell GPU (RTX 50xx, B100/B200 — compute capability >= 12.0):
docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-sm120

# Other NVIDIA GPU:
docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu

3. Verify

Call check_environment to verify everything is set up, then run_demo to test on the bundled sample images.

Bundled Demo Images

Requirements

• Docker with nvidia-container-toolkit
• NVIDIA GPU with drivers installed
• Python 3.10+ with mcp SDK (pip install mcp)

Security & Privacy

• Images are processed inside an ephemeral Docker container (--rm flag)
• The container has no network access beyond --network host (needed for GPU)
• No data leaves the host machine
• The container is destroyed immediately after each OCR run

External Endpoints

None. All processing is local.

Official References

• PaddleOCR-VL docs: https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PaddleOCR-VL.html
• Blackwell-specific: https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.html