Parse Document

Use the dedicated document tools as the default interface. Do not fall back to manual lit CLI commands unless the user explicitly asks for the raw command line workflow or the extension tools are unavailable.

Tool routing

• Use document_parse to extract text or JSON from a local document.
• Use document_search to find a phrase and get page numbers plus bounding boxes for citations/source locations.
• Use document_screenshot to render pages as PNG image blocks when visual layout matters.

Recommended workflow for known text: document_search first, then document_screenshot only for relevant pages.

Efficient parsing

Choose the smallest useful output

• Use format: "text" when the user wants to read, summarize, quote, search, or review the document.
• Use format: "json" when the user needs structured page data, text positions, or bounding boxes.
• Avoid JSON unless coordinates or programmatic structure matter.