All releases
ImprovedAI

AI apps can read PDFs, decks and large HTML files page by page

get_document and fetch now return a PDF's text by page, a PowerPoint deck's by slide, and a large HTML file's visible text in parts, a few pages per call, so an assistant can read a 40-page PDF to the end.

  • MCP server
  • PDFs and decks. get_document and fetch return the text under --- Page N --- (or --- Slide N --- for a .pptx, with speaker notes), about 40,000 characters per call. page_from and page_to (1-based, inclusive) choose the pages; total_pages and next_page_from say where to continue. Word and Excel files come back by section and sheet the same way.
  • Scanned PDFs say "No extractable text (looks scanned)" instead of returning nothing. OCR isn't available yet.
  • Large HTML files come back as their visible text (scripts and styles out, headings and link text kept) in parts: part, total_parts, next_part. Smaller HTML files still return their source unless you pass format: text.
  • Access is unchanged: a file you can't open returns nothing.