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

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 @xberg-io/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 html-to-markdown-cli

Language SDKs

pip install html-to-markdown                                # Python
npm install @xberg-io/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 XbergIo.HtmlToMarkdown               # C#
npm install @xberg-io/html-to-markdown-wasm                 # WASM

• Java (Maven): io.xberg:html-to-markdown
• Elixir: {:html_to_markdown, "~> 3.8"} 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 for conversion (FILE is positional; omit or use - for stdin); the only subcommand is mcp.
• 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_html and extract_metadata as agent tools, so an MCP client can convert an HTML string 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 xberg vs crawlberg

• 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.
• xberg — 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.
• crawlberg — 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 = crawlberg. Non-HTML documents = xberg.

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 "@xberg-io/html-to-markdown";

// Node's convert() returns a ConversionResult object directly.
const result = 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).

Configuration

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

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

html = "<h1>Title</h1><p>Body text.</p>"
result = convert(
    html,
    ConversionOptions(
        heading_style="atx",
        wrap=True,
        wrap_width=100,
        preprocessing=PreprocessingOptions(enabled=True, preset="aggressive"),
    ),
)

Metadata extraction

The library convert() extracts metadata by default; the CLI needs --json --extract-metadata (see the extracting-metadata skill). Fields include document (title, description, language, canonical_url, open_graph), headers, links (with link_type), images, and structured_data (JSON-LD/Microdata/RDFa).

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. This holds for Node.js too — convert() returns a ConversionResult object directly; do not JSON.parse() it.
2. --json outputs JSON, not Markdown. Omit --json for plain Markdown.
3. --include-structure, --extract-inline-images, and --no-content require --json.
4. The conversion CLI is flags-only. FILE is positional; the only subcommand is mcp (starts the MCP server).
5. --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