List PDFs in the configured literature folder (de-duplicated by SHA-256). Returns paper_id, file name, title candidate, page count, parse status. Never returns full text. Indexes not-yet-parsed PDFs on first call.

Page count, native PDF bookmarks (if any), inferred section headings with pages, per-page text coverage and quality flags, References location. Inferred items are marked inferred and may be wrong.

Text of a page range (1-based, max 10 pages per call), each page kept separate with its real page number. Output is bounded by max_chars; when truncated, 'next' tells you exactly where to continue (start_page + char_offset). PDF text layer only: verify formulas/tables with render_paper_page.

Render ONE page as a PNG image (for checking formulas, tables, figures, two-column reading order). scale 0.5–3 (default 1.5); longest edge capped at 2400 px.

Reference-list entries as printed (raw_text) with PDF page numbers, plus candidate fields (number/authors/year/title/venue/DOI) parsed by deterministic rules. Fields are unverified candidates (fields.year only when quality.year=verified_by_pattern; otherwise see candidates/issues); text left out as another article's (continuation-marker evidence) is listed in 'excluded' with its page. A citation says nothing about inheritance or experimental comparison.

Keyword search over page text (Chinese/English substring, whitespace- and width-insensitive; multiple space-separated terms must all occur on the same page). Returns paper_id + page + snippet; re-read hits with read_paper_pages. Query ≤ 200 chars.

Rescan the PDF folder and (re)build the local derived cache. Never modifies PDFs. force=true re-parses every paper; otherwise only new/changed files are parsed. One failing file does not stop the others.