html-to-markdown

html-to-markdown

html-to-markdown is a high-performance HTML→Markdown converter with a Rust core and 12 native language bindings. It converts HTML to CommonMark Markdown, Djot, or plain text in a single pass, optionally extracting metadata, tables, inline images, and a structured document tree.

Use this skill when writing code that:

• Converts HTML strings, files, or live URLs to Markdown, Djot, or plain text
• Extracts metadata (title, OG tags, JSON-LD/Microdata/RDFa, headers, links, images, language) from HTML
• Extracts structured table data (GFM markdown + cell grids) from HTML
• Extracts a structured document-structure tree
• Extracts inline images (data URIs, SVGs) from HTML
• Uses preprocessing to clean noisy HTML (ads, navigation, forms) before conversion

Capability map

Capability	CLI	SDKs
HTML→Markdown / Djot / plain text	html-to-markdown FILE	convert(html, options)
Read HTML from stdin / file / URL	cat f | …, FILE, --url URL	convert(htmlString, …)
~30 config options (headings, code blocks, lists, escaping, wrapping…)	flags	ConversionOptions
Metadata extraction	--json (default-extracted)	result.metadata
Table extraction	--json → tables[]	result.tables
Document structure tree	--json --include-structure	include_document_structure=true
Inline image extraction	--json --extract-inline-images	extract_images=true
HTML preprocessing	--preprocess [--preset …]	PreprocessingOptions
Extraction-only (no Markdown body)	--json --no-content	options + read fields

Installation

CLI

# (Homebrew 6.0+ requires explicit trust for third-party taps)
brew trust xberg-io/tap
brew install xberg-io/tap/html-to-markdown
# or run without a persistent install (the CLI proxy package self-installs the binary):
npx @kreuzberg/html-to-markdown-cli --help
uvx --from html-to-markdown-cli html-to-markdown --help
# or download a prebuilt binary from the latest GitHub release:
#   https://github.com/xberg-io/html-to-markdown/releases/latest
# or build from source:
cargo install --git https://github.com/xberg-io/html-to-markdown html-to-markdown-cli

Language SDKs

pip install html-to-markdown                                # Python
npm install @kreuzberg/html-to-markdown                      # TypeScript / Node.js
cargo add html-to-markdown-rs                                # Rust (features: metadata default; full = all)
gem install html-to-markdown                                 # Ruby
composer require xberg-io/html-to-markdown              # PHP
go get github.com/xberg-io/html-to-markdown/packages/go/v3   # Go
dotnet add package KreuzbergDev.HtmlToMarkdown               # C#
npm install @kreuzberg/html-to-markdown-wasm                 # WASM

• Java (Maven): dev.kreuzberg:html-to-markdown
• Elixir: {:html_to_markdown, "~> 3.6"} in mix.exs
• R: install.packages("htmltomarkdown", repos = "https://xberg-io.r-universe.dev")
• C (FFI): pre-built .so / .dll / .dylib from GitHub releases

CLI vs SDK — which to use

• CLI — one-shot conversions, shell pipelines, fetching a single URL, ad-hoc metadata/table extraction via --json | jq. Flags only, no subcommands. FILE is positional; omit or use - for stdin.
• SDK — embedding conversion in application code, batch processing, custom element conversion (visitor pattern, Rust), and tight loops where process spawn overhead matters.
• MCP server — html-to-markdown mcp exposes convert/extract as agent tools, so an MCP client can convert HTML directly with no shell-out. This plugin auto-registers it; see the using-the-mcp-server skill.

Both share the same ConversionResult shape, so output is interchangeable.

When to use html-to-markdown vs kreuzberg vs kreuzcrawl

• html-to-markdown — you already have HTML (a string, a file, or a single URL) and want clean Markdown plus structured metadata/tables. No OCR, no document parsing, no crawling.
• kreuzberg — you have documents (PDF, Office, images, email, archives) and need full text/table/metadata extraction with optional OCR. Use it when the input is not already HTML.
• kreuzcrawl — you need to crawl or scrape many pages, follow links, and handle JS-rendered sites with a headless-Chrome fallback. It uses html-to-markdown internally for the HTML→Markdown step.

Rule of thumb: single HTML in → Markdown out = html-to-markdown. Many URLs / a site = kreuzcrawl. Non-HTML documents = kreuzberg.

CLI quick start

# Convert a file to stdout
html-to-markdown input.html

# Convert and save
html-to-markdown input.html -o output.md

# Read from stdin
cat page.html | html-to-markdown

# Fetch and convert a URL
html-to-markdown --url https://example.com > out.md

# Full ConversionResult as JSON (content, tables, metadata, images, warnings)
html-to-markdown --json input.html

# JSON with document structure tree
html-to-markdown --json --include-structure input.html

# Extraction-only (no Markdown body)
html-to-markdown --json --no-content input.html

# Aggressive web-page cleanup
html-to-markdown input.html --preprocess --preset aggressive

SDK quick start

Rust

use html_to_markdown_rs::convert;

let result = convert("<h1>Hello World</h1><p>A paragraph.</p>", None)?;
println!("{}", result.content.unwrap_or_default());

Python

from html_to_markdown import convert

result = convert("<h1>Hello World</h1><p>A paragraph.</p>")
print(result.content)   # # Hello World\n\nA paragraph.
print(result.metadata)  # title, links, headers, …

TypeScript / Node.js

import { convert } from "@kreuzberg/html-to-markdown";

// Node's convert() returns a JSON string — always JSON.parse() it.
const result = JSON.parse(convert("<h1>Hello World</h1><p>A paragraph.</p>"));
console.log(result.content);

ConversionResult fields

All languages return the same structure (dict, object, or struct).

Field	Description
content	Converted text (Markdown/Djot/plain). null only in extraction-only mode.
metadata	Title, OG, headers, links, images, structured data.
tables	Tables with grid (structured cells) and markdown fields.
images	Extracted inline images (requires inline-image extraction).
document	Structured document tree when structure extraction is enabled.
warnings	Non-fatal processing warnings (message, kind).

Configuration

All languages expose the same ~30 options. See references/configuration.md for the complete table. Common ones:

Option	Values	Default
heading_style	atx, underlined, atx-closed	atx
code_block_style	backticks, indented, tildes	backticks
output_format	markdown, djot	markdown
wrap / wrap_width	bool / 20–500	off / 80
autolinks (SDK) / --no-autolinks (CLI)	bool / flag	true (on); disable in CLI with --no-autolinks
preprocessing	minimal / standard / aggressive	off

Rust (builder)

use html_to_markdown_rs::{convert, ConversionOptions, HeadingStyle, OutputFormat};

let options = ConversionOptions::builder()
    .heading_style(HeadingStyle::Atx)
    .output_format(OutputFormat::Markdown)
    .wrap(true)
    .wrap_width(100)
    .build();
let result = convert(html, Some(options))?;

Python (dataclass)

from html_to_markdown import convert, ConversionOptions, PreprocessingOptions

result = convert(
    html,
    ConversionOptions(heading_style="atx", wrap=True, wrap_width=100),
    PreprocessingOptions(enabled=True, preset="aggressive"),
)

Metadata extraction

Metadata is extracted by default; in the CLI it appears under metadata in --json output. Fields include document (title, description, language, charset, open_graph), headers, links (with link_type), images, and structured_data (JSON-LD/Microdata/RDFa). See the extracting-metadata skill for details.

Table extraction

Tables appear in result.tables, each with a pre-rendered markdown string and a structured cell grid. Markdown tables also appear inline in content. See the extracting-tables skill.

Document structure extraction

Enable structure extraction (--include-structure on the CLI, include_document_structure=true in SDKs) to get a semantic node tree under document. Node types include heading, paragraph, list, list_item, table, image, code, quote, group, metadata_block.

Common pitfalls

1. convert() returns a result object, not a string. Access .content for the Markdown text.
2. Node.js convert() returns a JSON string. Always JSON.parse(convert(html)) — NAPI-RS serializes the result for performance.
3. --json outputs JSON, not Markdown. Omit --json for plain Markdown.
4. --include-structure, --extract-inline-images, and --no-content require --json.
5. CLI has no subcommands. Everything is a flag; FILE is positional.
6. --preset, --keep-navigation, --keep-forms require --preprocess.

Additional resources

• CLI Reference — every flag, JSON shape, exit codes
• Configuration Reference — all 30+ options with defaults
• Rust API Reference — signatures, builder, feature flags
• Python API Reference — functions, dataclasses, type hints
• TypeScript API Reference — functions, interfaces, Buffer support
• Other Bindings — Go, Ruby, PHP, Java, C#, Elixir, R, WASM, C FFI

GitHub: https://github.com/xberg-io/html-to-markdown