The fastest path: copy one prompt, paste it into your favourite AI coding tool, and let it wire up the blog. Every agent gets the same pre-filled context.
What the dashboard gives you
Open Ship for any site. The page surfaces a single, comprehensive prompt with:
- Your three credentials (
MENTIONWELL_API_URL,MENTIONWELL_SITE_SLUG,MENTIONWELL_API_KEY) - The brand profile (audience, tone, CTA)
- The exact API endpoints to call
- File scaffolds for the agent to create
- Acceptance criteria the agent must verify before claiming "done"
Click Copy setup prompt. You're done with Mentionwell.
Where to paste it
| Tool | How |
|---|---|
| Cursor | Open your destination site repo. Cmd-L → paste. |
| Claude Code | In your destination site repo: claude → paste. |
| Lovable / Bolt.new / v0 | Click "Open with prompt" in the dashboard. The prompt is pre-encoded in the URL. |
| ChatGPT / Claude.ai | "Open in ChatGPT" / "Open in Claude" buttons load the prompt. |
| Replit Agent | New Repl with the prompt pasted as the initial brief. |
| Windsurf | Cascade chat → paste. |
Why this works
The prompt is built from your live site config. The agent doesn't have to guess at:
- env var names
- the brand voice
- which framework you're using
- whether you need ISR / SWR / cache headers
- what to do about JSON-LD / canonical / OG tags
…all of which it would otherwise hallucinate. We hand it the answer.
What the agent should produce
Every prompt ends with the same "definition of done":
BLOGOTO_*env vars added to.env.localand the hosting provider./blogindex renders posts./blog/[slug]renders an article (HTML, metadata, JSON-LD).- RSS
<link>added to<head>. - Build passes.
- Browser-tested both pages.
- Reply with
BLOGOTO READY.
If your agent skips any of those, ask it to verify before merging.
Going off-script
You don't have to use the prefilled prompt. You can also point your agent at this docs site directly:
Read https://mentionwell.com/llms.txt and follow the quickstart for the framework I'm using. Use these env vars: MENTIONWELL_API_URL=…
Both paths work. The prefilled prompt is just faster.