Connect your site to Claude and hand it the technical work: pages, images, menus, Elementor layouts, SEO, WooCommerce. 115 operations in the free plugin, every change previewed and reversible. Open source (GPL). Every write shows a diff before it lands, is verified against the live site after, and can be rolled back from an audit log in wp-admin. The agent never gets PHP, SQL or file access.
Hi Product Hunt! I do WordPress and SEO work for clients - 700+ orders on Fiverr over the years. When AI agents got good enough to do real site work, I hit a problem: there was no safe way to let one touch a live client site. Browser automation is fragile, and most plugins hand the agent PHP or SQL.
So I built SiteHelm: an MCP server that lives inside WordPress. The agent gets 115 typed operations - content, media, menus, Elementor, SEO, WooCommerce - and nothing else. No PHP, no SQL, no file access.
The safety model is the product:
- Every write is two calls: a preview that returns a diff, then an apply bound to that exact diff.
- The engine snapshots before writing and reads the site back after. A mismatch reports as a failure, not a success.
- Every change lands in an audit log in wp-admin, and applied changes roll back from there.
The core plugin is free and open source (GPL). Pro adds WooCommerce writes, Elementor Pro, plugin and theme management, and deeper SEO.
I would love to hear what would make you trust an AI agent on your own site. Ask me anything!