Installation

Requirements

• Node ≥ 20 or Bun ≥ 1.0
• ESM only: use import. From CommonJS, await import("@nalinor/mupdf4llm") (the mupdf dependency uses top-level await, which require() cannot load).
• The only runtime dependency is mupdf (Artifex's official WASM bindings). No native build, no system libs.
• Optional, only for OCR of table cells with the default RapidOCR engine: npm i ppu-paddle-ocr onnxruntime-node (ONNX Runtime is a native package of a few hundred MB).

Install

npm

npm i @nalinor/mupdf4llm

bun

bun add @nalinor/mupdf4llm

pnpm

pnpm add @nalinor/mupdf4llm

yarn

yarn add @nalinor/mupdf4llm

Optional: LlamaIndex adapter

If you want the PDFMarkdownReader integration, install llamaindex alongside — it's declared as an optional peer dependency:

npm i llamaindex

Without llamaindex, the adapter still loads — it just returns plain { text, metadata } objects with the same shape as a LlamaIndex Document (the metadata field is always metadata, never extra_info).

Optional: parity tests

To run the parity suite locally you also need Python with pymupdf4llm:

pip install pymupdf4llm

The tests spawn python3 -c "..." to compare the TS output against the upstream. CI sets this up automatically — only needed if you contribute to the library itself.

Verify

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

const md = await toMarkdown(readFileSync("any.pdf"));
console.log(md.slice(0, 80));

If you see Markdown coming out, you're set. Continue to the quick start.