pdf-analyst

v1.4.2 MCP-compatible documentsdata Apache-2.0 updated 2026-07-30

Parse PDFs into clean text blocks and tables. Falls back to OCR for scanned documents and attaches page-level citations to every extracted block, so downstream answers can point at sources.

Installation

$ skillbank install pdf-analyst@1.4.2
# OCR engine bundled · no system deps required

Configuration

# pdf-analyst.config.yaml
ocr:
  enabled: true
  language: "eng"        # tesseract language packs, e.g. eng+deu
  dpi: 200
tables:
  detect: true
  min_confidence: 0.6
citations:
  page_labels: true      # use printed page numbers when present

Inputs

Input	Type	Required	Description
`file`	string (path) / bytes	yes	PDF file to analyze. Max 200 MB per file.
`pages`	string	no	Page selection, e.g. `"1-5,8"`. Default: all pages.
`extract_tables`	boolean	no	Attempt table reconstruction. Default `true`.
`ocr`	"auto" \| "force" \| "off"	no	OCR strategy for pages with no text layer. Default `"auto"`.
`output`	"blocks" \| "markdown"	no	Structured blocks with citations, or flat markdown. Default `"blocks"`.

Version history

v1.4.22026-07-30
- Fixed table detection on two-column layouts (cells were merged across columns).
- Cut peak memory use of the OCR fallback path roughly in half via page-by-page processing.
v1.4.02026-06-15
- Page-level citations are now attached to every output block, including OCR'd pages.
- Added page_labels option to prefer printed page numbers over PDF indices.
v1.3.02026-04-20
- OCR fallback for scanned documents (Tesseract bundled, eng pack included).
- Fixed rotated-page text ordering for 90° and 270° rotations.
v1.0.02025-09-18
- Initial release: text block extraction, basic table detection, markdown output.

← back to catalog