llmstxt-standalone 0.1.1

Generate llms.txt from built HTML documentation

llmstxt-standalone

Generate /llms.txt, /llms-full.txt, and per-page markdown files from built HTML documentation, following the llms.txt spec.

This tool works on pre-built HTML, making it useful for environments that cannot run MkDocs plugins (e.g., Zensical) or when you want llms.txt generation as a separate build step. For standard MkDocs workflows, see mkdocs-llmstxt.

Installation

Requires Python 3.10+.

# Run without installing
uvx llmstxt-standalone

# Install as a CLI tool
uv tool install llmstxt-standalone  # or: pipx install

# Add to a project
uv add llmstxt-standalone  # or: pip install

Usage

# Run from project root (expects mkdocs.yml and site/)
llmstxt-standalone

# Explicit paths
llmstxt-standalone --config mkdocs.yml --site-dir ./build --output-dir ./dist

# Preview without writing files
llmstxt-standalone --dry-run

# Suppress output
llmstxt-standalone --quiet

# Show detailed progress
llmstxt-standalone --verbose

Option	Short	Default	Description
--config	-c	mkdocs.yml	Path to MkDocs config file
--site-dir	-s	site	Path to built HTML directory
--output-dir	-o	same as site-dir	Where to write output files
--dry-run	-n		Preview without writing
--quiet	-q		Suppress output
--verbose	-v		Show detailed progress
--version	-V		Show version

Output

The tool generates three outputs:

1. llms.txt — an index file with markdown links to all pages
2. llms-full.txt — concatenated content of all pages
3. Per-page .md files alongside the HTML

The per-page markdown files make the URLs in llms.txt resolve to actual content. If your site is at https://docs.example.com/, the URL https://docs.example.com/install/index.md returns markdown instead of HTML.

Configuration

The tool reads your mkdocs.yml for site metadata. You can configure llmstxt output explicitly or let it derive structure from your nav.

Explicit configuration

plugins:
  - llmstxt:
      markdown_description: |
        Extra context for LLMs about your project.
      full_output: llms-full.txt
      content_selector: article.md-content__inner
      sections:
        Getting Started:
          - index.md
          - install.md
        Usage:
          - guide/basics.md
          - guide/advanced.md

Option	Default	Description
markdown_description	""	Additional context for LLMs, appears after site description
full_output	llms-full.txt	Filename for concatenated content
content_selector	auto-detect	CSS selector for main content
sections	derived from nav	Section names mapped to page lists

Automatic fallback

Without an explicit llmstxt plugin config, sections derive from your nav structure. Top-level pages go into a "Pages" section; nested nav items become sections named by their keys.

MkDocs settings

The tool respects use_directory_urls from your mkdocs.yml. When enabled (the default), install.md maps to install/index.md; when disabled, it maps to install.md.

Content extraction

If content_selector is not set, the tool tries these selectors in order:

1. .md-content__inner (Material for MkDocs)
2. [role="main"] (default MkDocs theme)
3. article
4. main
5. The entire document

Title resolution

Page titles resolve in this order:

1. The title from your nav structure
2. The HTML <title> tag (with site name suffix stripped)
3. The first <h1> tag
4. A title derived from the filename

Programmatic use

from pathlib import Path
from llmstxt_standalone.config import load_config
from llmstxt_standalone.generate import generate_llms_txt

config = load_config(Path("mkdocs.yml"))
result = generate_llms_txt(config, site_dir=Path("site"))

print(result.llms_txt)       # Index content
print(result.llms_full_txt)  # Full content
print(result.markdown_files) # List of written .md paths

Compatibility

• Produces output identical to mkdocs-llmstxt when configured the same way
• Handles Unicode, international characters, and special characters
• Works with Material for MkDocs, ReadTheDocs, and the default MkDocs theme
• Parses configs containing Python YAML tags like !python/object/apply

License

MIT