WordPress.com Opens the Door to AI Agents: Claude, ChatGPT, and Cursor Can Now Manage Your Site
What Happened
On March 20, 2026, Automattic announced that AI agents can now take direct action on WordPress.com sites through natural conversation. Agents like Claude, ChatGPT, and Cursor can draft blog posts, edit pages, manage comments, update metadata, and organize content with tags and categories — all through the Model Context Protocol (MCP).
The integration works through WordPress.com's MCP server, which acts as a bridge between AI assistants and your site. When you connect an AI agent, it can see your site's content structure and take actions on your behalf — similar to giving a virtual assistant access to your dashboard, but through conversation instead of clicking.
Key details:
- Supported agents — Claude (Anthropic), ChatGPT (OpenAI), Cursor, and any MCP-compatible tool
- Capabilities — Create posts, edit pages, publish content, manage comments, update metadata, organize tags and categories
- Safety — Opt-in only, nothing enabled by default. Users choose which capabilities are active, site by site. Posts written by AI are saved as drafts by default.
- Availability — All WordPress.com paid plans. Enable at
wordpress.com/me/mcp - Scale — This puts AI agents to work on a platform where 70 million new posts are published every month
Why It Matters
This is a significant step in WordPress's Phase 3 (Collaboration) roadmap. While WordPress 7.0 focuses on human-to-human collaboration (real-time co-editing), this announcement adds human-to-AI collaboration to the mix.
For content creators: you can now ask Claude to "write a draft about our spring sale" or tell ChatGPT to "clean up the comments on yesterday's post" — and it happens directly on your WordPress site without copy-pasting or switching tabs.
For the broader WordPress ecosystem: this positions WordPress.com as one of the first major CMS platforms with native AI agent integration. Competitors like Wix and Squarespace have built-in AI writers, but WordPress.com's approach is different — it opens the platform to any MCP-compatible agent rather than building a proprietary AI tool.
This also connects directly to the Abilities API being developed for WordPress 7.0 (self-hosted). While today's announcement is for WordPress.com specifically, the same MCP protocol will be available for self-hosted WordPress.org sites through the WordPress MCP Adapter plugin.
What You Should Do
- If you use WordPress.com — Check out the MCP settings at
wordpress.com/me/mcpto see the available capabilities. Start with read-only access to test how AI agents interact with your content before enabling write permissions. - If you use self-hosted WordPress — The MCP Adapter plugin is already available on GitHub. WordPress 7.0 (April 9) will integrate these capabilities more deeply into core.
- Keep AI drafts as drafts — AI-generated content should always be reviewed before publishing. The default draft behavior is there for a reason.
Sources
Written by Marvin
Our team tests and reviews WordPress products to help beginners make confident choices.
Learn more about our team →You might also like
WordPress Playground Gets MCP Server: Let AI Coding Agents Build WordPress Sites in Your Browser
A new @wp-playground/mcp package lets AI coding agents like Claude and Cursor interact with WordPress Playground directly — reading files, executing PHP, and building sites through conversation.
postCritical WooCommerce Vulnerability Patched: CSRF Flaw Could Create Rogue Admin Accounts
A critical CSRF vulnerability affecting 52 WooCommerce versions (5.4–10.5.2) could let attackers create admin accounts and access customer data. Auto-patches rolled out March 2.
postWordPress 7.0 RC1 Delayed: Real-Time Collaboration Switched Off by Default, Client-Side Media Pulled
WordPress 7.0 Release Candidate 1 delayed to March 24. Client-side media processing pulled from the release entirely, real-time collaboration switched off by default.
postWordPress Ships 3 Security Patches in 24 Hours After 6.9.2 Breaks Sites
WordPress released versions 6.9.2, 6.9.3, and 6.9.4 within 24 hours after the initial security patch caused white-screen crashes and left vulnerabilities incompletely fixed.