unmhtml

Convert MHTML files — or any HTML document with its resources — to a single standalone HTML string: CSS inlined, images and fonts embedded as data URIs, active content sanitized for safe display.

pip install unmhtml

Convert an MHTML archive:

from pathlib import Path

from unmhtml import load_mhtml, to_standalone_html

document = load_mhtml(Path("saved_page.mhtml").read_bytes())
html = to_standalone_html(document)

Or bring your own HTML and resources — a rendered page plus the bodies captured in a HAR archive, for example:

from unmhtml import Document, Resource, to_standalone_html

document = Document(
    html=rendered_html,
    resources={
        "https://example.com/style.css": Resource(css_bytes, "text/css"),
        "https://example.com/logo.png": Resource(png_bytes, "image/png"),
    },
    base_url="https://example.com/page.html",
)
html = to_standalone_html(document)

References resolve against base_url and match resource keys exactly; CSS references inside a stylesheet resolve against the stylesheet's own URL. Anything unresolvable is neutralized, so the result never makes network requests.

Conversion is secure by default: scripts, event handlers and dangerous URLs are removed, forms and meta redirects are defused, and CSS that could still fetch or execute is stripped. Each neutralization has its own flag on Security (remove_javascript, disable_forms, remove_meta_redirects, sanitize_css), all defaulting to on:

from unmhtml import Security, to_standalone_html

html = to_standalone_html(
    document,
    security=Security(remove_javascript=False),  # flags are independent
)

The contract is reduction, not isolation: the output makes no network requests, but this is defense in depth, not a sandbox — display untrusted output in a sandboxed context of your own, such as a sandboxed iframe without allow-scripts, combined with a Content-Security-Policy.

• load_mhtml(data) -> Document — parse MHTML content; raises ValueError when no HTML part exists
• Document(html, resources, base_url) — a parsed document; resources map absolute URLs to Resource values
• Resource(data, mime_type) — one resource; the MIME type is guessed from the reference URL when omitted
• to_standalone_html(document, *, security=Security()) -> str — the document as one self-contained HTML string

The output is an HTML fragment: html/head/body wrappers are removed and the original doctype is re-emitted. HTML comments are removed, and every anchor carries rel="noopener noreferrer".

Python 3.8+. The single dependency is nh3 (Python bindings to the Rust ammonia sanitizer) — zero transitive runtime dependencies, wheels for all major platforms.

MIT