readpdf

Convert PDFs to text for token-efficient AI agent reading.

Why

Reading PDFs with vision costs 100–500 tokens per page. Converting to text first lets any AI agent read only the sections it needs — saving up to 90% of tokens.

Install

pip install readpdf

Requires pdftotext:

# macOS
brew install poppler

# Linux
apt install poppler-utils

Usage

# Convert to file, then read selectively
readpdf paper.pdf -o paper.txt

# Extract a single page
readpdf paper.pdf -o paper.txt -p 3

# Extract a page range
readpdf paper.pdf -o paper.txt --pages 3-7

# Print to stdout
readpdf paper.pdf

Works with any AI agent that has shell access: Claude, GPT-4, Gemini, Cursor, etc.

How it works

Without readpdf:
  PDF → AI vision → ~200 tokens/page × 30 pages = 6,000 tokens (full doc, every time)

With readpdf:
  PDF → pdftotext → paper.txt (on disk, 0 tokens)
  AI reads only offset/limit chunks it needs → ~300 tokens total

Step 1. readpdf paper.pdf -o paper.txt runs pdftotext locally — no AI tokens consumed.

Step 2. The AI uses a file-reading tool (e.g. Read with offset/limit) to load only the relevant lines. Because the text file already exists on disk, the AI never pays the cost of processing the entire PDF.

Why not MCP? MCP tool results return the full content back into the AI's context window — same cost as reading directly. A disk file lets the AI pull exactly the slice it needs, nothing more.

License

MIT