html2any

Compile messy HTML into compact context for AI agents.

Library

While building websites, people may need to render rich text into different formats. For example, I've got an <video> tag, but I wanna render it with my own React video component. But I also want to render the whole html easily rather than parse it manually.

html2any helps you render an HTML string. It parses HTML and gives you the ability to transform it from the source format into your target format.

API

html2any provides the following APIs:

• AST(Object) parse(String source)

• void transform(AST ast, function rule)

• void html2any(html, function rule)

• parse

Build an AST from HTML/XML source.

• transform

Convert the AST to a final form with a custom rule.

• html2any

Convert the html/xml to the final form directly.

Usage

npm i -S html2any

import html2any, { parse, transform } from 'html2any'

const html = escapeHTMLEntity(`<div>123</div>`)

const ast = parse(html)[0]

function rule(node, children) {
  if (typeof node === 'string') {
    return node
  } else {
    return <div>{node}</div>
  }
}

const vdom = transform(ast, rule)
// JSX vdom form of html
// { type: 'div', props: {...}, children: '...' }

Or you can just call html2any directly

const vdom = html2any(html, rule)

CLI

Convert a local file, URL, or stdin into compact Markdown:

html2any md <file|url|->

# Short alias
html2 md <file|url|->

Examples:

html2 md page.html > page.context.md
cat page.html | html2 md - --url https://example.com/page

The Markdown output keeps docs-friendly structure: title, description, headings, paragraphs, lists, tables, code blocks, and links.

The extractor is deterministic and token-size friendly. It drops scripts, styles, hidden content, and layout chrome while preserving semantically useful docs content.

How It Works

Use html2any to construct an AST from an HTML string, then convert each node recursively with the rule passed to the transform function.

For example, translate a <p> tag into a React Native component like <Text style={styles.paragraph}> with prepared styles, then decode the paragraph content to avoid HTML entity issues.