Launched originally in October last year to allow AI to access statistics and content from websites, support for the MCP (Model Context Protocol) protocol on WordPress.com is evolving.
This update introduces 19 new capabilities spread across six types of content: articles, pages, comments, categories, tags, and media. The goal is to transform AI, from a simple drafting assistant, into a true collaborator capable of manipulating the site’s structure.
Automattic ensures that this opening is supervised by a strict security layer. The system inherits existing WordPress permissions natively. A user with a “Contributor” role will never be able to publish content via AI, even if technically capable. In principle, each action must be explicitly validated by a human before being executed on the database.
Productivity: AI becomes “Design-Aware”
Beyond just text input, these AI agents are now able to analyze the active theme of the site, including color palettes, fonts, and “block patterns.”
This contextual understanding also allows AI to generate landing pages or “about” sections that adhere to the graphic charter without manual intervention from the integrator.
It also becomes possible to automate time-consuming tasks via a chat interface:
- Generation and structuring of categories and subcategories
- Audit and automatic correction of accessibility (Alt tags of images, metadata)
- Moderation of comments with contextual responses
- Configuration of internal meshes based on tags suggested by AI
“Your AI agent discovers the available operations, determines what is necessary, and guides you through the process – confirming each step before making changes,” says Jonathan Bossenger of WordPress.com.
Security and control: granular deployment
Aware of the risks associated with automation, WordPress.com requires that each content creation be initially saved as a “Draft.”
Furthermore, any deletion action is reversible for 30 days via the trash, except for categories and tags, for which AI requires double confirmation from the user.
Finally, the management of these functions is done through a dedicated dashboard, where administrators can enable or disable each writing capability in a granular manner. This approach allows testing automation on specific segments (e.g., media management) before extending it to the overall structure of the site.
