commit_message.txt

feat: Add MCP server with PDF extraction and enhance hybrid method

Major enhancements:
- Add MCP (Model Context Protocol) server on port 8002 with 4 tools:
  - describe_image_with_openai: OpenAI GPT-4 Vision image analysis
  - describe_image_with_ollama: Local Ollama vision model analysis
  - describe_image_with_claude: Anthropic Claude vision analysis
  - extract_pdf_content: PDF content extraction with hybrid method support

- Implement hybrid PDF extraction method combining text_and_images + page_as_image:
  - Preserves exact text while adding visual descriptions
  - Optimized prompts to prevent LLM paraphrasing
  - Parallel processing for better performance
  - Separate output directories to prevent file collisions

- Infrastructure improvements:
  - Add unified server management script (run_servers.sh)
  - Update Docker configurations with poppler-utils for PDF processing
  - Set OpenAI default model to gpt-4o across all endpoints
  - Add comprehensive MCP setup documentation and examples

- Bug fixes:
  - Fix filename collision in PDF extractors
  - Resolve temporary directory cleanup issues
  - Fix API key environment variable fallback

The hybrid PDF extraction method is now the recommended default, providing
the best balance of text accuracy and visual analysis.