Triage a local PDF without extracting its text (typically 10-50ms). Returns whether the document is text-based, scanned, image-based or mixed, a confidence score, and the exact 1-indexed pages that need OCR. ALWAYS call this before pdf_extract on an unfamiliar or large document: it is cheap, and it tells you whether local extraction is worth attempting at all. If it reports scanned, image_based, or encoding issues, do not extract — route the document to an OCR service instead.

Extract a local PDF to Markdown, preserving headings, lists and tables. Output is TRUNCATED by default to protect your context window — to read a long document, call repeatedly with the pages parameter rather than raising maxChars. Call pdf_classify first on unfamiliar documents. If the response carries a critical warning (encoding issues, no text layer, right-to-left script), the text is unreliable and must not be quoted as fact.

Find text inside a local PDF and return matching pages with surrounding context. Much cheaper than pdf_extract when you only need to locate something — use this first on long documents, then pdf_extract with the pages it reports.

Return only the tables from a local PDF as Markdown, skipping prose. Useful for invoices, financial statements and reports where the numbers are the point. Tables are detected from the PDF's own drawing operations and text alignment — the cell values are read directly from the document, not guessed by a model or OCR.