Claude, ChatGPT, Cursor, and Other AI Agents Can Now Take Direct Action on WordPress.com Sites Through Natural Conversation
WordPress.com powers 43% of all websites on the internet. Now it is handing that traffic to AI agents.
The platform announced Friday that its MCP server can now let AI agents write, edit, and publish content on customer sites — along with managing comments, organizing tags and categories, fixing alt text and metadata, and handling structural site changes. The numbers that put this in context: 20 billion pageviews and 409 million unique visitors per month, across a network that hosts more than four in ten websites on the internet.
The capability is opt-in. Users toggle on specific permissions at wordpress.com/mcp, then connect their preferred AI client — Claude, Cursor, ChatGPT, or any other MCP-enabled tool — and direct the agent using natural language. All changes require user approval. AI-authored posts are saved as drafts by default before any publishing step. The Activity Log tracks everything.
This follows the introduction of MCP read-only support last fall. The Model Context Protocol, built by Anthropic, is an open standard — not a proprietary interface — which means WordPress.com's adoption of it as a platform is significant for the protocol ecosystem. WordPress.com supporting read access was a signal. WordPress.com supporting write access, five months later, is the follow-through.
What is being enabled is not fully autonomous publishing. It is a specific kind of delegation: a human directing an agent to handle the mechanics of content creation while retaining approval at every step. The agent drafts, the user approves, the content goes live. The OAuth 2.1 auth model is the infrastructure that makes explicit consent the default rather than the exception.
The broader context is the question of what the web looks like when platforms at this scale hand content production to agents. Meta recently acquired Moltbook, a social network where AI agents posted, replied, and connected with one another. Anthropic has experimented with an AI writing its own blog under human oversight. WordPress.com's approach — requiring approval at every step, tracking everything in the Activity Log — is more conservative than those experiments, but the scale is not conservative at all. Twenty billion pageviews a month is not a test environment.
The plugin and hosting ecosystem built around WordPress has historically assumed human operators at the dashboard. If agents can handle content creation, moderation, and site organization directly, the economic logic of many WordPress tools changes. That is a problem for a large ecosystem of auxiliary products, even if it is not a problem for the platform itself.
Source: TechCrunch and WordPress.com blog