AI Agents Can Now Run Your WordPress Site
▼ Summary
– Automattic has added write capabilities to WordPress.com’s MCP integration, allowing AI agents like Claude to create and modify site content through natural conversation, with human approval required for every action.
– The update introduces 19 new operations across six content types, enabling agents to draft posts, build pages, manage comments, and reorganize categories from a single prompt.
– A key safety feature is explicit human approval, where agents describe planned changes for confirmation, new posts default to drafts, and deletions are sent to a recoverable trash.
– The AI agents are theme-aware, meaning they can read a site’s design system and generate content that matches its specifications, such as colors and block patterns.
– These write capabilities are available on all WordPress.com paid plans, work with compatible MCP clients like Claude and ChatGPT, and enforce existing user role permissions.
The landscape of website management is shifting with the introduction of AI-powered write capabilities on WordPress.com. Automattic has significantly upgraded its Model Context Protocol (MCP) integration, moving beyond simple data queries to allow AI agents like Claude and ChatGPT to actively create and manage site content. This evolution transforms these tools from passive observers into active assistants, capable of drafting posts, building pages, and handling comments—all initiated through natural conversation and executed only with explicit human approval at every step.
For the last half-year, connecting an AI to a WordPress.com site offered a view-only experience. You could ask about analytics or identify outdated content, but the agent couldn’t make any changes. The recent update fundamentally alters that dynamic by adding a suite of nineteen new operations across six core content types. Now, from a single prompt, an authorized agent can draft a blog post, construct a landing page using your theme’s design patterns, manage comment moderation, reorganize your category hierarchy, or even audit and fix missing alt text in your media library.
This functionality is built upon the MCP framework, an open protocol that standardizes how applications provide context to large language models. Initially launched in a read-only capacity, the system has been progressively enhanced. Following updates for secure OAuth 2.1 authentication and an official Claude Connector, these new write permissions represent the logical culmination of that development path.
A core tenet of the system is user control and oversight. The AI agent must describe its intended action and seek confirmation before creating, updating, or deleting anything. New posts are saved as drafts by default, allowing for human review before publication. Any attempt to modify a live post triggers a clear warning that changes will be public immediately. For deletions, items like posts or comments are sent to the trash, where they remain recoverable for thirty days. The system fully respects WordPress user roles, ensuring an Editor cannot alter site settings and a Contributor cannot publish without approval.
A particularly sophisticated feature is theme awareness. Before generating content, the AI can analyze the site’s design system—including colors, fonts, and spacing—to produce drafts that seamlessly inherit the existing visual identity. This allows for the creation of on-brand pages and posts that fit naturally within the site’s established aesthetic.
These new capabilities are available immediately on all WordPress.com paid plans. Users activate them through a dedicated MCP dashboard, where they can granularly toggle specific permissions for each connected site. The update works with a range of compatible clients, including Claude, ChatGPT, and Cursor. Given WordPress’s dominant market share, powering over 43% of all websites, the potential scale of this AI-assisted content management is substantial.
The broader MCP ecosystem continues to grow. An adapter for self-hosted WordPress sites is progressing toward inclusion in core software, and Automattic is implementing similar protocols across its other products like WooCommerce. This trend points toward standardized AI agent access becoming a foundational element of software architecture.
For site owners, the central consideration is trust. Granting an AI the ability to modify a live website is a more significant step than using it for analysis. Automattic addresses this directly by making the mandatory approval model the cornerstone of the feature and providing detailed, operation-level controls as the default setup, placing final editorial authority firmly in human hands.
(Source: The Next Web)
