WordPress.com has expanded its MCP integration from read-only to read-write. AI agents connected through Claude, ChatGPT, Cursor, or any MCP-enabled tool can now draft and publish posts, build pages, manage comments, organize categories and tags, and update media metadata — all through natural conversation.
The update adds 19 new write capabilities across six content types. This builds on the MCP support WordPress.com launched last October, which gave agents a window into site content, analytics, and settings but stopped short of letting them make changes.
What Agents Can Do Now
- Draft and publish blog posts with categories, tags, and meta descriptions
- Build and update pages using the site's existing block patterns and design specs
- Manage comments — approve, reply, or clean up without opening the dashboard
- Organize content by creating, renaming, and restructuring categories and tags
- Fix media metadata — alt text, captions, and titles for accessibility and SEO
Design-Aware Output
Before creating content, the agent can query the active theme's colors, fonts, spacing, and block patterns. Pages and posts inherit the site's design system automatically and adapt when you switch themes — a meaningful step beyond generic AI-generated markup.
Safety by Default
Every change requires explicit user approval. New posts default to draft status. Deletions go to trash (recoverable for 30 days) where supported, and the agent warns when deletion is permanent. All activity is logged in the site's Activity Log, and WordPress role permissions are enforced — an Editor can create posts but can't change site settings.
This matters because WordPress.com powers over 43% of all websites. When the dominant CMS gives AI agents write access with sensible guardrails, it sets the pattern for how every platform will handle agent permissions. The toggle-per-operation approach — enable only what you need, per site — is the right model for teams that want to experiment without handing over full control.