NameetP/pdfmux

What's New

Graphical PDF Detection

pdfmux now detects image-heavy PDFs (pitch decks, infographics, slides) and routes them to OCR or LLM extractors instead of fast text extraction. Previously, these PDFs would silently produce incomplete output.

Honest Confidence Scoring

Confidence scores now reflect actual extraction quality. A graphical PDF where image content was missed will score ~55% instead of falsely claiming 100%. Color-coded output: green (≥80%), yellow (50-79%), red (<50%).

Actionable Warnings

When extraction is limited, pdfmux tells you exactly what to do:

⚠ 35 of 47 pages contain images with text that could not be extracted.
  Install pdfmux[ocr] or pdfmux[llm] for better results.

MCP Server Quality Metadata

AI agents now receive confidence scores and warnings alongside extracted text, so they can make informed decisions about output quality.

Spaced-Text Cleanup

Fixes a common PDF artifact where text renders as W i t h o v e r → cleaned to With over.

FastExtractor Fallback

When pymupdf4llm returns empty (certain PDF encodings), pdfmux now falls back to raw fitz text extraction instead of producing empty output.