Reliable PDF Text Extraction

Problem

The

read_file

filetype='pdf'

• Return binary image data instead of text
• Fail with errors on certain PDF structures
• Lose formatting or structured content

Solution

Use

run_shell

read_file

Methods

Method 1: pdftotext (Recommended)

pdftotext input.pdf output.txt

Or to extract to stdout:

pdftotext input.pdf -

With layout preservation:

pdftotext -layout input.pdf output.txt

Method 2: pdfinfo (Metadata)

pdfinfo input.pdf

Useful for checking page count, dimensions, and PDF properties before extraction.

Method 3: Python with PyMuPDF (fitz)

import fitz  # PyMuPDF

doc = fitz.open("input.pdf")
text = ""
for page in doc:
    text += page.get_text()
doc.close()

Method 4: Python with pdfplumber (Tables)

import pdfplumber

with pdfplumber.open("input.pdf") as pdf:
    for page in pdf.pages:
        text = page.extract_text()
        tables = page.extract_tables()

Workflow

1. Check PDF exists and is readable:

   pdfinfo input.pdf 2>/dev/null || echo "PDF not accessible"
   

2. Extract text using pdftotext:

   pdftotext -layout input.pdf - > extracted_text.txt
   

3. If pdftotext fails, try Python fallback:

   import fitz
   doc = fitz.open("input.pdf")
   for i, page in enumerate(doc):
       print(f"--- Page {i+1} ---")
       print(page.get_text())
   doc.close()
   

4. Verify extraction succeeded:

   • Check output is non-empty
   • Verify text is readable (not binary/garbled)
   • Confirm expected content is present

When to Use

pdftotext

pdftotext -layout

PyMuPDF

pdfplumber

Example Integration

# In your agent workflow, prefer this pattern:
result = run_shell(command="pdftotext document.pdf -", timeout=30)
if result.stdout and len(result.stdout.strip()) > 0:
    content = result.stdout
else:
    # Fallback to Python
    content = execute_python_to_extract_pdf("document.pdf")

Notes

• Install tools if needed:

  apt-get install poppler-utils

  (for pdftotext/pdfinfo)
• Python libraries:

  pip install pymupdf pdfplumber

• Some PDFs are image-scanned and require OCR (Tesseract) instead
• Always validate extracted content before proceeding with downstream tasks