Publishing

Refine can write the article that gets a brand cited, then put it on the brand’s own site. Through the MCP server, an assistant runs that loop on request; a scheduled agent runs it on its own. All of it requires the Autopilot plan.

Three ways onto the site#

Anywhere else (another CMS, a static site, a Reddit thread, LinkedIn), take the HTML from get_content, publish it yourself, then call mark_content_posted with the URL. That closes the item in the queue and starts measuring its effect on visibility.

On request: the five-call loop#

What an assistant does when someone asks “write and publish an article on our weakest prompt”:

1. get_opportunities        → the prompt losing the most demand
2. generate_content         → { contentId }, generation runs in the background
3. get_content  (poll)      → status "completed", HTML body, meta, JSON-LD
4. list_publish_targets     → is a WordPress or Webflow site connected?
5. publish_article          → the article lands in the CMS (draft by default)

On a schedule: the agent loop#

An agent that runs every few days needs three calls, and Refine keeps the pace for it: next_article_to_publish refuses to hand over an article until minGapDays (2 to 5, default 3) have passed since the last post. Scheduling the agent more often than that is safe; extra runs simply stop at cooldown.

every run:
  next_article_to_publish
    ready: true             → publish article.id:
                                 publish_to_site            (site runs @getrefine/next)
                              or publish it yourself, then
                                 mark_content_posted { contentId, publishedUrl }
    reason: "cooldown"      → stop; try again after next_eligible_at
    reason: "queue_empty"   → get_opportunities → generate_content
                              (the next run will find it)

Articles come out oldest first. The response carries everything a publisher needs: title, html, meta_description, json_ld and the prompt it answers. Hermes Agent ships with this loop as the refine-publisher skill: see the quickstart to install it.

Publishing on a Next.js site#

@getrefine/next renders Refine articles server-side on the customer’s own domain: real HTML that AI crawlers, which do not run JavaScript, can read and cite. No CMS, no webhook, no rebuild.

Install

npm install @getrefine/next

Requires Next.js 14 or later (App Router) and React 18 or later. Copy the site key from Settings → Site publishing (one key per brand, starting with pk_) into the environment:

REFINE_SITE_KEY=pk_…            # Refine → Settings → Site publishing
# optional
REFINE_REVALIDATE=300           # seconds between refreshes (ISR)

Add the article route

// app/blog/[slug]/page.tsx
export {
  generateStaticParams,
  generateMetadata,
  default,
} from "@getrefine/next/blog";

That single file serves every published article at /blog/<slug>, with its title and meta description in the page metadata. The markup is <article class="refine-article"> with an h1, a time and a .refine-article-body, ready to style.

Add an index (optional)

The package does not create the /blog listing. Build it with getArticles(), or render articles your own way with getArticle(slug):

// app/blog/page.tsx
import { getArticles } from "@getrefine/next";

export default async function BlogIndex() {
  const articles = await getArticles();
  return (
    <ul>
      {articles.map((a) => (
        <li key={a.slug}>
          <a href={`/blog/${a.slug}`}>{a.title}</a>
        </li>
      ))}
    </ul>
  );
}

getArticles() returns { slug, title, description, publishedAt, updatedAt }[] newest first; getArticle(slug) adds html and returns null for an unknown slug. Both read the public content API.