Quick start

Five-minute tour of the two main entry points and a few common options.

toMarkdown(buf, opts?)

Returns a single Markdown string for the whole document.

import { toMarkdown } from "@nalinor/mupdf4llm";
import { readFileSync } from "node:fs";

const md = await toMarkdown(readFileSync("paper.pdf"));

On Bun:

const md = await toMarkdown(await Bun.file("paper.pdf").bytes());

Browser:

const buf = await fetch("paper.pdf").then((r) => r.arrayBuffer());
const md = await toMarkdown(buf);

toMarkdownPages(buf, opts?)

Returns one record per page. Useful for chunked RAG, where you want each chunk to carry the page number / TOC items as metadata.

import { toMarkdownPages } from "@nalinor/mupdf4llm";

const chunks = await toMarkdownPages(readFileSync("paper.pdf"));
for (const c of chunks) {
  console.log(`page ${c.metadata.page}: ${c.text.length} chars`);
}

Common combinations

Selected pages only:

await toMarkdown(buf, { pages: [0, 4, 5] });

Embed images inline as base64 data URLs:

await toMarkdown(buf, { embedImages: true });

Or save them to a folder and link by path:

await toMarkdown(buf, {
  writeImages: true,
  imagePath: "out/images",
  imageFormat: "jpg",
  dpi: 200,
});

Per-word coordinates inside the chunks:

await toMarkdownPages(buf, { extractWords: true });
// chunks[0].words[i] = { x0, y0, x1, y1, text, block, line, word }

Page separators in the joined output:

await toMarkdown(buf, { pageSeparators: true });

Disable table detection if it gets in the way:

await toMarkdown(buf, { tableStrategy: null });

Next

• Full options reference
• Tables guide — lines_strict vs lines vs text vs explicit
• LlamaIndex adapter