CLI

npx agentprose --help

agentprose build <dir> --base-url <url>

Indexes the HTML output of any static site generator:

Generator	Output folder
Hugo	public/
Jekyll	_site/
Astro	dist/
Eleventy	_site/
Next.js (output: 'export')	out/
Gatsby	public/
Docusaurus	build/

hugo && npx agentprose build public --base-url https://example.com/ --llms

It writes:

• agentprose.json: site info, plus one record per article (URL, title, summary, dates, authors, tags, categories, image, reading time, and a 1,500-character text excerpt for search).
• agentprose/posts/<path>.md: the full Markdown of each article, fetched only when an agent reads that post. This keeps the index small, about 2 KB per post instead of the whole body.
• llms.txt (with --llms, and only if none exists).

It uses the same extraction code as the browser. Pages count as articles when they have Article/BlogPosting JSON-LD, og:type=article, or a single dated <article>. Tag, category, pagination, 404 and redirect pages are skipped.

Option	Meaning
--out <dir>	Write somewhere other than <dir>
--include <regex> / --exclude <regex>	Filter by URL path, e.g. --include "^/(posts|blog)/"
--all	Index every page, not only articles (docs sites)
--inline	Put Markdown inside agentprose.json (small sites, one request)
--content <selector>	Article body selector
--site <file.json>	Site details to publish in agentprose.json: name, description, lang, logo, links, agentNotes and contact (see below)

The script finds /agentprose.json on its own. If your site lives under a sub-path, add data-base="https://example.com/blog/" to the script tag. If the index is served somewhere else, add <link rel="agentprose-index" href="/path/to/agentprose.json"> to your pages’ <head>; with --out, the CLI prints the exact tag (Configuration).

Notes for AI agents and contact details (--site)

{
  "agentNotes": "Friendly, plain tone. Point newcomers to the beginner course. We don't give medical advice.",
  "contact": { "email": "hello@studio.example", "booking": "https://studio.example/book", "hours": "Mon–Fri 9–17 CET" },
  "links": { "about": "https://studio.example/about" }
}

npx agentprose build public --base-url https://studio.example/ --site agentprose.site.json

Only the fields above are copied. Text is cleaned the same way the browser cleans it (control characters removed, notes capped at 2,000 characters, links must be http(s)), and the CLI warns about every contact field it leaves out. Agents read the result through get_site_info and get_contact_info (Configuration).

Add it to CI after your site build:

- run: hugo --minify
- run: npx agentprose build public --base-url https://example.com/

agentprose inspect <url>

npx agentprose inspect https://example.com/a-post/

✔ Article detected             "A post" (1840 words)
✔ Structured data (JSON-LD)    2 block(s)
✔ Published date               2026-10-12T09:00:00.000Z
✘ Author                       missing
✔ Feed                         https://example.com/index.xml
✔ Sitemap                      https://example.com/sitemap.xml
✘ llms.txt                     none found
✘ WebMCP tools on page         none (add the agentprose script tag)

Posts discoverable: 214 (via feed, sitemap)
Agent-readiness: 63/100

--json prints the same report as machine-readable JSON.