Clean content mode — strip subscription forms, related-post blocks, promotional banners and other theme/plugin injections from machine documents by rendering article content through WordPress core filters only. Off by default and fully reversible, with an mde_content_html filter for custom control.

AI discovery manifest (Pro) at /.well-known/ai-content.json — a compact machine-readable description of your site's content interface: publisher, canonical, URL patterns and discovery endpoints. A WP Markdown Endpoint convention, clearly declared as such in the document itself — not an industry standard.

Feed items now always carry a summary (the hand-written excerpt, or a clean trim of the content), plus authors, featured image and tags — an agent can triage the feed without fetching fifty articles.

llms.txt lists the 30 most recent documents instead of 200 and points at the Markdown sitemap and AI feed for the rest — a small discovery document, not an index.

HTML entities no longer leak into JSON taxonomy fields ("Oil & Gas"); root-relative link and image URLs are absolutised; and two settings saves within the same second now both invalidate the output cache.

Structured JSON endpoint (Pro) — every document now also answers on .json: title, publisher, author, ISO-8601 published/modified timestamps, canonical/markdown/json URLs, categories, tags, excerpt, featured image and the Markdown body. Fields that don't exist are omitted, never fabricated.

AI news feed (Pro) — a JSON Feed 1.1 of your latest content at /feed/ai/ (application/feed+json) and /feed/ai.json (application/json). Each item links its canonical page, .md and .json versions, with timestamps, summary, tags and category.

Machine-format discovery — eligible HTML articles now advertise their .md and .json twins with link rel="alternate" tags and matching Link response headers. Toggleable, on by default, included in free.

Front matter upgraded to the agent-publishing shape — publisher, ISO-8601 published and modified timestamps, primary category, an optional format label (e.g. "news") and language, alongside the existing fields. The 1.0 date field is kept so nothing parsing it breaks.

llms.txt now documents the whole machine interface — the .md/.json URL patterns, the AI feed, RSS and sitemaps — instead of only listing documents.

Delivery rebuilt for scale: machine documents are now publicly cacheable (Cache-Control, ETag, Last-Modified, 304 conditional requests) instead of forcibly uncacheable, and carry X-Robots-Tag: noindex with a canonical Link header — so CDNs cache them and they can never compete with your HTML page in search.

Cache invalidation now also fires on scheduled publishing, trash and every status transition, and clears the .json and feed caches alongside the Markdown.

YAML front matter could emit invalid YAML when a title contained a line break.

First release: .md endpoint for posts and pages, toggleable metadata lines, author display override, lightweight HTML-to-Markdown converter with fenced code blocks, live-preview settings screen. Pro: custom post types, per-post disable, YAML front matter, excerpt + featured image, table conversion, output cache, Markdown sitemap, llms.txt manifest.