Markfly

Markfly is a Python library that converts HTML into clean, readable Markdown. It walks an lxml.html element tree and emits CommonMark / GitHub-Flavored Markdown (GFM), correctly tracking links, images, lists, tables, and reference-style link definitions along the way.

Markfly is a Python port of jina-ai/reader's MarkifyService (originally written in TypeScript), rebuilt on top of lxml for use in Python projects.

Features

• CommonMark + GFM output — headings, paragraphs, emphasis, lists, links, images, code blocks, blockquotes, and (optionally) GFM tables, strikethrough, and checkboxes.
• Smart image resolution — falls back through srcset, data-src, data-lazy-src, data-original, and sibling <picture>/<source> elements when src is missing, empty, or a placeholder.
• Reference-style links — supports inlined, referenced (full / collapsed / shortcut), and discarded link styles.
• Configurable formatting — heading style (ATX or Setext), bullet markers, code fence style, emphasis/strong delimiters, and more.
• Custom rules — register your own per-tag replacement rules, or mark specific tags to be kept as raw HTML.
• MathML support — converts MathML to LaTeX via an optional pluggable math_converter callable, with sensible fallback when one isn't provided.
• Base URL resolution — automatically resolves relative links and image sources against a baseUrl.

Installation

Markfly isn't published to PyPI yet. In the meantime, drop markfly.py into your project, or install it from your local checkout:

pip install lxml

Once published:

pip install markfly

Quick start

from markfly import html_to_markdown

html = """
<h1>Hello <em>World</em></h1>
<p>This is <strong>bold</strong> and a <a href="https://example.com">link</a>.</p>
<ul>
  <li>one</li>
  <li>two</li>
</ul>
"""

print(html_to_markdown(html))

Output:

# Hello _World_

This is **bold** and a [link](https://example.com).

*   one
*   two

Usage

Simple conversion

The fastest path is the html_to_markdown() convenience function, which parses the HTML string and returns Markdown in one call:

from markfly import html_to_markdown

markdown = html_to_markdown("<p>Hello <b>world</b></p>")

Pass options as keyword arguments:

markdown = html_to_markdown(
    html,
    gfm=True,
    headingStyle="atx",
    baseUrl="https://example.com",
)

Using MarkflyService directly

For more control — reusing an instance across many conversions, registering custom rules, or converting an already-parsed lxml element — use MarkflyService directly:

from lxml.html import fromstring
from markfly import MarkflyService, MarkflyOptions

options = MarkflyOptions(gfm=True, codeBlockStyle="fenced")
service = MarkflyService(options)

root = fromstring("<h2>Title</h2><p>Body text.</p>")
markdown = service.markfly(root)

Note: each call to service.markfly(root) resets internal state (link/image tracking, list and table stacks), so a single MarkflyService instance can safely be reused for multiple, independent conversions.

Options

All options are set via MarkflyOptions, passed either as an options= object or as keyword arguments to html_to_markdown().

GFM mode

Pass gfm=True to enable GitHub-Flavored Markdown extensions:

html = """
<table>
  <tr><th>Name</th><th>Role</th></tr>
  <tr><td>Ada</td><td>Engineer</td></tr>
</table>
<p>Status: <s>Pending</s> Done</p>
<input type="checkbox" checked> Ship it
"""

print(html_to_markdown(html, gfm=True))

Output:

| Name | Role |
| --- | --- |
| Ada | Engineer |

Status: ~~Pending~~ Done

- [x] Ship it

GFM mode also enables MathML → LaTeX conversion for <math> elements (see Math support below).

Link styles

html = '<a href="https://example.com">Example</a>'

# Inlined (default)
html_to_markdown(html)
# -> [Example](https://example.com)

# Referenced, full style
html_to_markdown(html, linkStyle="referenced", linkReferenceStyle="full")
# -> [Example][1]
# ->
# -> [1]: https://example.com

# Discarded — keeps the text, drops the link
html_to_markdown(html, linkStyle="discarded")
# -> Example

Resolving relative URLs

Set baseUrl to resolve relative href and src values against a real origin:

html = '<a href="/docs">Docs</a> <img src="/logo.png" alt="logo">'
html_to_markdown(html, baseUrl="https://example.com")

[Docs](https://example.com/docs) ![logo](https://example.com/logo.png)

Image fallback resolution

Markfly doesn't just read src. When src is missing, empty, or a known placeholder (e.g. a base64 GIF/PNG spacer), it tries, in order:

1. srcset / data-srcset — picks the highest-resolution candidate
2. data-src, data-lazy-src, data-original
3. Sibling <source> elements inside a <picture> wrapper

This makes it resilient against lazy-loaded images from real-world scraped pages.

Custom rules

Register your own conversion logic for specific tags with addRule:

from lxml.html import fromstring
from markfly import MarkflyService, MarkflyRule

service = MarkflyService()

def render_mark(text, element, options, service):
    return f"=={text}=="

service.addRule("highlight", MarkflyRule(filter="mark", replacement=render_mark))

root = fromstring("<p>This is <mark>important</mark>.</p>")
print(service.markfly(root))

Or preserve specific tags as raw HTML instead of converting them:

service.keep("iframe")

Math support

MathML → LaTeX conversion requires a converter callable, since there's no drop-in PyPI equivalent to @nomagick/mathml-to-latex. Without one, Markfly falls back to the element's alttext attribute or its plain text content.

from markfly import MarkflyService, MarkflyOptions

def my_math_converter(mathml_string: str) -> str:
    # plug in your own MathML -> LaTeX conversion here
    ...

service = MarkflyService(MarkflyOptions(gfm=True), math_converter=my_math_converter)

API reference

html_to_markdown(html, options=None, **kwargs) -> str

Convenience entry point. Parses an HTML string and returns Markdown.

MarkflyService(options=None, math_converter=None)

The main converter class.

• .markfly(element) -> str — convert an lxml.html element tree to Markdown.
• .addRule(name, rule) — register a custom per-tag replacement rule.
• .keep(tag) — preserve a tag as raw HTML instead of converting it.
• .use(rule_fns) — apply a list of rule-registration functions.

MarkflyOptions

Dataclass holding all converter options (see Options above).

MarkflyRule(filter, replacement)

Dataclass describing a custom rule: filter is a tag name or list of tag names; replacement is a callable (text, element, options, service) -> str.

Notes & known limitations

• MathML conversion has no built-in LaTeX backend — supply your own math_converter.
• blob: and data: URLs are never used as a baseUrl, since resolving relative links against them doesn't make sense.
• preformattedCode and footnoteStyle are present in MarkflyOptions for API parity with the original TypeScript implementation but aren't fully wired up yet.

Credits

Markfly is a Python port of the Markdown conversion logic from jina-ai/reader.

License

Add your license of choice here before publishing to PyPI.