kepano/defuddle

de·​fud·dle /diˈfʌdl/ transitive verb
to remove unnecessary elements from a web page, and make it easily readable.

Beware! Defuddle is very much a work in progress!

Defuddle extracts the main content from web pages. It cleans up web pages by removing clutter like comments, sidebars, headers, footers, and other non-essential elements, leaving only the primary content.

Overview

Defuddle takes a URL or HTML, finds the main content, and returns cleaned HTML or Markdown. Defuddle was created for the browser extension Obsidian Web Clipper, but it is designed to run in any environment.

Defuddle can be used as a replacement for Mozilla Readability with a few differences:

• More forgiving, removes fewer uncertain elements.
• Provides a consistent output for footnotes, math, code blocks, etc.
• Uses a page's mobile styles to guess at unnecessary elements.
• Extracts more metadata from the page, including schema.org data.

Usage

Browser

import Defuddle from 'defuddle';

// Parse the current document
const defuddle = new Defuddle(document);
const result = defuddle.parse();

// Access the content and metadata
console.log(result.content);
console.log(result.title);
console.log(result.author);

Node.js

defuddle/node accepts a DOM Document from any implementation (JSDOM, linkedom, happy-dom, etc.).

import { parseHTML } from 'linkedom';
import { Defuddle } from 'defuddle/node';

const { document } = parseHTML(html);
const result = await Defuddle(document, 'https://example.com/article', {
  markdown: true
});

console.log(result.content);
console.log(result.title);
console.log(result.author);

Or with JSDOM:

import { JSDOM } from 'jsdom';
import { Defuddle } from 'defuddle/node';

const dom = new JSDOM(html, { url: 'https://example.com/article' });
const result = await Defuddle(dom.window.document, 'https://example.com/article');

Note: for defuddle/node to import properly, the module format in your package.json has to be set to { "type": "module" }

CLI

Defuddle includes a command-line interface for parsing web pages directly from the terminal. You can run it with npx or install it globally. The CLI accepts a file path, a URL, or HTML piped over stdin.

# Parse a local HTML file
npx defuddle parse page.html

# Parse a URL
npx defuddle parse https://example.com/article

# Parse HTML from stdin
cat page.html | npx defuddle parse

# Parse fetched HTML from stdin as markdown
curl -L https://stephango.com/saw | npx defuddle parse --markdown

# Output as markdown
npx defuddle parse page.html --markdown

# Output as JSON with metadata
npx defuddle parse page.html --json

# Extract a specific property
npx defuddle parse page.html --property title

# Save output to a file
npx defuddle parse page.html --output result.html

# Enable debug mode
npx defuddle parse page.html --debug

# Use a custom User-Agent (helps with sites that return 403 to the default UA)
npx defuddle parse https://example.com/article --user-agent "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15"

CLI Options

When no <source> argument is provided, defuddle parse reads HTML from stdin. You can also pass - explicitly to force stdin input.

Installation

npm install defuddle

For Node.js usage, install a DOM implementation:

npm install linkedom

Or use JSDOM:

npm install jsdom

CLI installation

To use the defuddle command globally, install it with the -g flag:

npm install -g defuddle

Or use npx to run the CLI without installing globally:

npx defuddle parse https://example.com/article

Response

Defuddle returns an object with the following properties:

Bundles

Defuddle is available in three different bundles:

1. Core bundle (defuddle): The main bundle for browser usage. No dependencies.
2. Full bundle (defuddle/full): Includes additional features for math equation parsing and Markdown conversion.
3. Node.js bundle (defuddle/node): For Node.js environments. Accepts any DOM Document (e.g. from linkedom, JSDOM, or happy-dom). Includes full capabilities for math and Markdown conversion.

The core bundle is recommended for most use cases. It still handles math content, but doesn't include fallbacks for converting between MathML and LaTeX formats. The full bundle adds the ability to create reliable <math> elements using mathml-to-latex and temml libraries.

Options

HTML standardization

Defuddle attempts to standardize HTML elements to provide a consistent input for subsequent manipulation such as conversion to Markdown.

Headings

• The first H1 or H2 heading is removed if it matches the title.
• H1s are converted to H2s.
• Anchor links in H1 to H6 elements are removed and become plain headings.

Code blocks

Code block are standardized. If present, line numbers and syntax highlighting are removed, but the language is retained and added as a data attribute and class.

<pre>
  <code data-lang="js" class="language-js">
    // code
  </code>
</pre>

Footnotes

Inline references and footnotes are converted to a standard format:

Inline reference<sup id="fnref:1"><a href="#fn:1">1</a></sup>.

<div id="footnotes">
  <ol>
    <li class="footnote" id="fn:1">
      <p>
        Footnote content.&nbsp;<a href="#fnref:1" class="footnote-backref">↩</a>
      </p>
    </li>
    </ol>
</div>

Math

Math elements, including MathJax and KaTeX, are converted to standard MathML:

<math xmlns="http://www.w3.org/1998/Math/MathML" display="inline" data-latex="a \neq 0">
  <mi>a</mi>
  <mo>≠</mo>
  <mn>0</mn>
</math>

Callouts

Callout and alert elements from various sources are standardized to blockquotes with a data-callout attribute. When converting to Markdown, these become Obsidian-style callouts.

Supported sources:

• GitHub markdown alerts (div.markdown-alert)
• Obsidian callouts (div.callout[data-callout])
• Callout asides (aside.callout-*)
• Bootstrap alerts (div.alert.alert-*)

The standardized HTML follows the Obsidian Publish format:

<div data-callout="info" class="callout">
  <div class="callout-title">
    <div class="callout-title-inner">Info</div>
  </div>
  <div class="callout-content">
    <p>This is an informational callout.</p>
  </div>
</div>

In Markdown:

> [!info] Info
> This is an informational callout.

Development

Build

To build the package, you'll need Node.js and npm installed. Then run:

# Install dependencies
npm install

# Clean and build
npm run build

Third-party services

When using parseAsync(), if no content can be extracted from the local HTML, Defuddle may fetch content from third-party APIs as a fallback. This only happens when the page HTML contains no usable content (e.g. client-side rendered SPAs). You can disable this by setting useAsync: false in options.

• FxTwitter API — Used to extract X (Twitter) article content, which is not available in server-rendered HTML.

Debugging

Debug mode

You can enable debug mode by passing an options object when creating a new Defuddle instance:

const result = new Defuddle(document, { debug: true }).parse();

// Access debug info
console.log(result.debug.contentSelector); // CSS selector path of chosen main content element
console.log(result.debug.removals);        // Array of removed elements with reasons
console.log(result.debug.errors);          // Errors caught during extraction, if any

When debug mode is enabled:

• Returns a debug field in the response with detailed information about content extraction
• More verbose console logging about the parsing process
• Preserves HTML class and id attributes that are normally stripped
• Retains all data-* attributes
• Skips div flattening to preserve document structure

The debug field contains:

Each removal entry contains:

Pipeline toggles

You can disable individual pipeline steps to diagnose content extraction issues:

// Skip content scoring to see if it's removing content incorrectly
const result = new Defuddle(document, { removeLowScoring: false }).parse();

// Skip hidden element removal (useful for CSS sidenote layouts)
const result = new Defuddle(document, { removeHiddenElements: false }).parse();

// Skip small image removal
const result = new Defuddle(document, { removeSmallImages: false }).parse();

Content selector

Use contentSelector to bypass Defuddle's auto-detection and specify the main content element directly:

const result = new Defuddle(document, {
  contentSelector: 'article.post-content'
}).parse();

If the selector doesn't match any element, Defuddle falls back to auto-detection.