kreuzberg-dev Extracting metadata
Use this when the user wants structured metadata out of HTML rather than (or in addition to) the Markdown body — page title, description, language, Open Graph tags, structured data, the heading outline, links, or image references.
Metadata lives in result.metadata and is surfaced on the CLI through
--json. On the CLI, metadata extraction is opt-in: pass
--extract-metadata alongside --json, otherwise result.metadata comes back
empty (document.title is null, headers/links/images/structured_data
are []). The library convert() call extracts metadata by default
(extract_metadata=True) — that default is a property of the API, not the CLI.
There are no per-field extraction flags: --extract-metadata populates all
sub-fields below at once.
Get all metadata
html-to-markdown --json --extract-metadata input.html | jq '.metadata'
# Extraction-only (skip the Markdown body)
html-to-markdown --json --extract-metadata --no-content input.html | jq '.metadata'
Metadata sub-fields
{
"metadata": {
"document": { "title": "...", "description": "...", "language": "en", "open_graph": {"title": "..."} },
"headers": [ { "level": 1, "text": "Main Heading" } ],
"links": [ { "href": "https://example.com", "link_type": "external" } ],
"images": [ { "src": "photo.jpg", "alt": "A photo", "image_type": "external" } ],
"structured_data": [ /* JSON-LD, Microdata, RDFa blocks */ ]
}
}
Metadata flag
There is one metadata flag: --extract-metadata. With --json set it populates
all sub-fields above (document, headers, links, images,
structured_data) under result.metadata — select what you need with jq.
| Flag | Effect |
|---|---|
--extract-metadata |
With --json: populate result.metadata. In plain-text mode (no --json): prepend title + meta tags as a YAML frontmatter block (----delimited) at the top of the Markdown output |
# Pull just the document-level metadata and the heading outline
html-to-markdown --json --extract-metadata --no-content input.html \
| jq '{title: .metadata.document.title, lang: .metadata.document.language, outline: [.metadata.headers[].text]}'
Common queries
# Title + canonical language
html-to-markdown --json --extract-metadata input.html | jq '{title: .metadata.document.title, lang: .metadata.document.language}'
# External links only
html-to-markdown --json --extract-metadata input.html | jq '[.metadata.links[] | select(.link_type == "external") | .href]'
# Open Graph card
html-to-markdown --json --extract-metadata input.html | jq '.metadata.document.open_graph'
# JSON-LD blocks
html-to-markdown --json --extract-metadata input.html | jq '.metadata.structured_data'
Programmatic access
from html_to_markdown import convert
html = """
<html lang="en">
<head><title>My Article</title></head>
<body>
<h1>Main Heading</h1>
<a href="https://example.com">External link</a>
<img src="photo.jpg" alt="A photo">
</body>
</html>
"""
result = convert(html)
meta = result.metadata
print(meta.document.title) # "My Article"
print(meta.document.language) # "en"
print(meta.headers[0].text) # "Main Heading"
print(meta.links[0].link_type) # "external"
print(meta.images[0].alt) # "A photo"
Metadata is available from the single convert() call in Python, Rust, Go,
Ruby, and Elixir; in TypeScript read it off the returned result object. In Rust
it requires the metadata feature (on by default).
See ../html-to-markdown/references/cli-reference.md (Metadata section) and
../html-to-markdown/references/configuration.md for field details.