WordPress.com allows AI agents to create, publish, and oversee your website.

      Automattic has introduced write functionalities to WordPress.com’s MCP integration, allowing AI agents like Claude and ChatGPT to generate posts, create pages, manage comments, and reorganize content through natural language interactions, with human oversight at every stage.

      For the majority of the last six months, connecting an AI agent to your WordPress.com site meant granting it limited access; you could ask Claude or ChatGPT questions about your content, access site analytics, or find posts that hadn’t been modified in a year—helpful but essentially passive interaction.

      However, on Friday, Automattic provided a more interactive option.

      WordPress.com has rolled out write features for its Model Context Protocol (MCP) integration, permitting AI agents to directly create and alter content on your site. The update introduces 19 new operations across six types of content: posts, pages, comments, categories, tags, and media. With a single natural language command, an agent can draft and publish a post, design a landing page using your theme’s block patterns, approve and respond to comments, reorganize categories, or add missing alt text throughout your media library.

      The MCP architecture, an open standard that streamlines how applications provide context to large language models, was first launched on WordPress.com in October 2025, initially offering read-only access to agents, allowing them to query your site but not make modifications.

      A second update in January 2026 introduced OAuth 2.1 authentication, simplifying secure connections for AI clients. In February, Automattic released an official Claude Connector, which was also read-only at that time. Today’s addition of write capabilities represents the culmination of ongoing platform development.

      This feature is structured to prioritize explicit human approval. Before creating, updating, or deleting any content, the agent clearly outlines its intended actions and requests confirmation. New posts are automatically saved as drafts, allowing users to review them before publishing, while modifying a published post triggers a warning indicating that changes will be instantly visible. Deleting posts, pages, comments, and media moves items to the trash, where they can be restored within 30 days. Deletions of categories and tags, which cannot be trashed by WordPress, come with an additional confirmation warning indicating that such deletions are irreversible. Every action taken is recorded in the site’s Activity Log.

      User role permissions are fully respected: an Editor can create and edit posts but lacks the authority to modify site settings, while a Contributor can draft content but cannot publish it.

      A particularly interesting technical feature of this implementation is theme awareness. Prior to creating a post or page, the agent can access the site’s design system, including colors, fonts, spacing, and block patterns, and produce content that adheres to these specifications.

      These write capabilities are now accessible on all paid WordPress.com plans. Users can activate them via the MCP dashboard at wordpress.com/me/mcp by enabling the specific operations they wish to allow on each site.

      Compatible clients include Claude, ChatGPT, Cursor, and any other MCP-enabled applications. WordPress.com boasts a significant share of the web; as noted in figures shared at Automattic’s State of the Word event in December 2025, WordPress powers over 43% of all websites globally and has a 60.5% market share in content management systems. The magnitude of write-capable AI agents operating within this framework is substantial.

      The MCP ecosystem is rapidly growing. The WordPress MCP Adapter, which offers similar functionality on self-hosted WordPress sites, is progressing toward inclusion in WordPress Core.

      Other Automattic products, such as WooCommerce and Beeper, have their specific MCP implementations. This trend of standardized AI agent access to application functionalities, rather than isolated integrations, is evolving into a foundational architectural principle rather than just an experimental phase.

      For users of WordPress.com, a key concern is trust. Granting an AI agent write access to a live site poses a different challenge compared to merely asking it to summarize traffic data. Automattic has clearly recognized this concern, emphasizing the approval model as a central aspect of the announcement and setting granular per-operation controls as the default configuration.