markitdown-ts

markitdown-ts converts various file types to Markdown without filesystem, process, stream, or any other runtime-specific dependencies. It can run in browsers, workers, edge functions, and all other modern JavaScript environments.

Supported formats

• Plain text, Markdown, CSV, and JSON
• HTML
• XML, RSS, and Atom feeds
• Jupyter notebooks (.ipynb)
• Word documents (.docx)
• Excel workbooks (.xlsx)
• PowerPoint presentations (.pptx)
• PDF documents (.pdf)
• ZIP archives (.zip), with nested files converted recursively
• Wikipedia, YouTube metadata, and Bing result pages supplied as HTML

Audio, image metadata, local-path, and YouTube transcript features from the original project are intentionally left out because their implementations depended on runtime-specific modules, binaries, streams, or filesystem access. They may be added back over time as alternative implementations are found. Want to see one added? Contributions are welcome!

Installation

pnpm install @ryangarber/markitdown-ts

Usage

import { MarkItDown } from "@ryangarber/markitdown-ts";

const bytes: Uint8Array = /* plain bytes */;

const result = await new MarkItDown().convert(bytes);

console.log(result?.markdown);

The same convert method accepts:

• Uint8Array
• ArrayBuffer and SharedArrayBuffer
• DataView or any other typed array
• Blob and File
• Response

The file type is detected from, in order: the extension option, a File's name or a Content-Disposition filename, the URL's path (the url option and a Response's URL), a Blob's type or a Content-Type header, the content's magic number, and finally the content itself (HTML, XML, RSS, Atom, notebooks, and plain text). Each candidate is tried until one converts, so pass extension only when the content is ambiguous.

const file = new File([docxBytes], "document.docx");
const result = await new MarkItDown().convert(file);

URLs

HTTP and HTTPS URLs are fetched with the global fetch. Pass a custom implementation when needed.

const result = await new MarkItDown().convert("https://example.com/article.html");

Browser requests remain subject to the remote server's CORS policy.

Responses

const response = await fetch("https://example.com/feed.xml");
const result = await new MarkItDown().convert(response);

API

type ByteSource = Uint8Array | ArrayBufferLike | ArrayBufferView | Blob;
type ConversionSource = ByteSource | Response | URL | string;

class MarkItDown {
  convert(source: ConversionSource, options?: ConverterOptions): Promise<ConverterResult>;

  /** @deprecated Pass the source directly to convert(). */
  convertBuffer(
    source: ByteSource,
    options: ConverterOptions & { file_extension: string }
  ): Promise<ConverterResult>;
}

convertBuffer remains as a compatibility alias, but convert(bytes, options) is preferred.

CLI

The markitdown command converts a file and prints its Markdown:

npx @ryangarber/markitdown-ts report.docx
markitdown --extension html page    # when the type can't be detected
markitdown report.pdf --json        # print { title, markdown }

To use it without Node, download the standalone build for your system (markitdown-darwin-arm64, markitdown-darwin-x64, markitdown-linux-arm64, markitdown-linux-x64, markitdown-win32-arm64.exe or markitdown-win32-x64.exe) from the latest release, or build it from source with pnpm build:cli (needs Bun).

Development

pnpm install
pnpm lint
pnpm typecheck
pnpm test --run
pnpm build        # library, CLI and man page
pnpm build:cli    # standalone binaries (needs Bun)

Biome handles both formatting and linting. tsdown produces ESM, CommonJS, source maps, and declarations in dist. Releases are cut by rylease from conventional commits: pushes to main dry-run the release, and dispatching the release workflow on main publishes to npm and attaches the CLI binaries to the GitHub release.

License

MIT License © 2026 Ryan Garber
MIT License © 2024 Vaibhav Raj