How to Extract Text from PDFs Using PyMuPDF

What this does

Structured text extraction from PDF documents is the ingestion starting point for most RAG pipelines. PyMuPDF provides fast, layout-aware text extraction with page-level granularity and metadata access. This guide demonstrates extracting readable text from a PDF file ready for embedding and vector database indexing.

Steps

1. Install PyMuPDF.

   pip install pymupdf --quiet
   

2. Open the PDF and iterate over pages.

   import fitz
   
   doc = fitz.open("document.pdf")
   print(f"Pages: {doc.page_count}")
   print(f"Metadata: {doc.metadata}")
   
   for page_num in range(doc.page_count):
       page = doc[page_num]
       text = page.get_text("text")
       print(f"\n--- Page {page_num + 1} ---")
       print(text[:200])
   doc.close()
   

3. Extract text with layout preservation.

   doc = fitz.open("document.pdf")
   page = doc[0]
   blocks = page.get_text("blocks")
   for block in blocks:
       x0, y0, x1, y1, content, block_type, *_ = block
       if block_type == 0:
           print(f"Y={y0:.0f}: {content[:80].replace(chr(10), ' ')}")
   doc.close()
   

4. Stream results for pipeline ingestion.

   import fitz, json
   
   doc = fitz.open("document.pdf")
   records = []
   for page_num in range(doc.page_count):
       page = doc[page_num]
       text = page.get_text("text").strip()
       if text:
           records.append({"source": "document.pdf", "page": page_num + 1, "content": text})
   doc.close()
   
   with open("extracted.jsonl", "w") as f:
       for rec in records:
           f.write(json.dumps(rec) + "\n")
   print(f"Wrote {len(records)} page records")
   

Verification

python3 -c "import fitz; print(f'PyMuPDF version: {fitz.__version__}')"
# Expected: PyMuPDF version: <version-number>

Common failures

• Scanned PDFs contain no extractable text. Use OCR libraries like Tesseract for scanned documents.
• Unicode rendering errors. Pass extracted text through unicodedata.normalize("NFKC").
• Memory exhaustion on large PDFs. Use page-level iteration instead of loading the entire document at once.
• Table extraction misaligned. Adjust coordinates manually for edge cases the detector misses.
• Hidden metadata in extracted text. Filter blocks by type and skip annotations with zero height.
• Version mismatch - The installed package or runtime differs from the command shown; check the version first and rerun the smallest verification command.
• Local environment drift - Another service, virtual environment, model, or path is being used; print the active binary path and configuration before changing the guide steps.

Related guides

• preprocess-text-vector-ingestion
• extract-tables-pdfs-structured-data