Options reference

Every option on MarkdownOptions. Defaults match pymupdf4llm.helpers.pymupdf_rag.to_markdown.

Page selection

Key	Type	Default	Notes
pages	number[]	all pages	0-based indices.
margins	number | [t, b] | [l, t, r, b]	0	Crop the page before extraction.
filename	string	""	Surfaced in PageChunk.metadata.file_path and used in image filenames.

Output mode

Key	Type	Default	Notes
pageSeparators	boolean	false	Insert --- end of page=N --- markers in joined output.

Use toMarkdownPages instead of toMarkdown when you want per-page chunks — the choice of return shape isn't an option, it's a separate entry point.

Text and styling

Key	Type	Default	Notes
ignoreCode	boolean	false	Suppress fenced code blocks for monospaced spans.
forceText	boolean	true	Emit text even on top of images.
hdrInfo	IdentifyHeaders | TocHeaders | function | false	auto IdentifyHeaders	See Headers.

Tables

Key	Type	Default	Notes
tableStrategy	"lines_strict" | "lines" | "text" | "explicit" | "pixels" | null	"lines_strict"	See Tables. null disables detection.
explicitTableGrids	{ hLines: number[]; vLines: number[] }[]	[]	For "explicit" only.
textSource	"pdf" | "ocr" | "auto"	"auto"	"auto". See OCR.
ocr	OcrEngine	RapidOCR	Needs ppu-paddle-ocr + onnxruntime-node unless you pass one.
ocrDpi	number	300	Render resolution for "pixels" and OCR.
detectOrientation	boolean	true	With "pixels", turn a page without a text layer scanned sideways. See Page rotation.

Images

Key	Type	Default	Notes
writeImages	boolean	false	Save each detected image region to imagePath.
embedImages	boolean	false	Inline images as data: URIs in the Markdown.
imagePath	string	""	Output directory for writeImages. Created if missing.
imageFormat	"png" | "jpg" | "jpeg"	"png"	Encoding format.
dpi	number	150	Rasterization DPI.
imageSizeLimit	number	0.05	Skip images smaller than this fraction of a page edge.

Words / chunks

Key	Type	Default	Notes
extractWords	boolean	false	Populate PageChunk.words with per-word bboxes.

Page rotation

Key	Type	Default	Notes
removeRotation	boolean	true	Derotate the page (visual-preserving) before processing, like PyMuPDF's remove_rotation().

Filtering

Key	Type	Default	Notes
fontsizeLimit	number	undefined	Skip spans whose font size is below this (in pt). Mirrors upstream FONTSIZE_LIMIT.
ignoreAlpha	boolean	false	Accept invisible text. No-op today — mupdf.js doesn't expose per-char alpha; field reserved for upstream parity.

Element selection

Key	Type	Default	Notes
elements	MarkdownElement[]	all (omitted)	Whitelist of markdown constructs to emit. Anything not listed becomes plain text.

When elements is omitted, every construct is emitted (the original behavior). When provided, only the listed constructs are produced; everything else falls back to plain text. Valid values:

Value	Controls
"bold"	**…** around bold spans
"italic"	_…_ around italic spans
"inlineCode"	`…` around monospaced spans (outside code blocks)
"codeBlock"	``` fenced blocks for monospaced lines
"header"	#-prefixed headings
"bulletList"	- bullet-list conversion
"link"	[text](url) hyperlinks
"table"	Markdown tables
"image"	 image references
"lineBreak"	<br> between wrapped lines inside table cells

elements combines with the legacy toggles (ignoreCode, hdrInfo: false, tableStrategy: null, writeImages/embedImages): a construct is emitted only if both the whitelist allows it and no legacy option disabled it.

// Plain text, headers and tables — but no bold/italic and no <br> in cells.
await toMarkdown(buf, {
  elements: ["header", "table", "link", "bulletList", "italic"],
});

// Strip all markup — just text.
await toMarkdown(buf, { elements: [] });

Misc

Key	Type	Default	Notes
showProgress	boolean	false	Render ProgressBar to stderr while iterating pages.

For the canonical types, see src/helpers/types.ts.