WordPress 7.0’s AI Stack Is Ready. The Release Isn’t.

On March 31, 2026, the WordPress core team did something unusual: they pulled the 7.0 release out of Release Candidate and paused pre-releases entirely. RC-stage delays are rare. They almost never happen for a single feature. This one did.

The culprit is Real-Time Collaboration. The current RTC implementation stores sync data in post_meta on an internal wp_sync_storage post type, and that approach disables WordPress’s persistent post query caches whenever an editor is open. It’s a performance regression serious enough to warrant a rewrite into a dedicated database table — the kind of change you don’t land with a late-cycle patch.

Pre-releases are paused until April 17. A new schedule is promised by April 22. Most community estimates now point to a mid-to-late May ship.

That’s the bad news. Here’s the interesting part.

The AI stack is ready. It’s just trapped in the release.

WordPress 7.0’s biggest non-RTC headline is the native AI infrastructure: the Abilities API, the MCP Adapter, the AI Client and Connectors API, and a JavaScript counterpart for browser-side agents. All of it is finished. All of it was tested against the RC. All of it is sitting on the shelf waiting for a collaboration feature to be rewritten.

RTC and the AI stack have nothing to do with each other. They share a release train, not a dependency. Every site that was planning to wire Claude Code, Cursor, or a custom agent into their WordPress install on launch day is now waiting on a feature they may not even use.

What “waiting six more weeks” actually costs

If your plan was to light up an agent the moment 7.0 shipped, the delay is not free:

• Six to eight weeks of agent-driven workflow. For sites that publish weekly or buy ads against new content, that’s real revenue on the table — content calendars unshipped, audits unrun, support tickets handled by hand.
• Real schedule risk. RC-stage delays are rare precisely because problems caught this late tend to expose more problems underneath. A second slip is not the base case, but it’s not zero either.
• A harder migration later. Teams that wait until 7.0 ships will start their Abilities API and MCP learning curve under launch-day pressure, at the same time every other site is configuring Connectors for the first time.

You already have an MCP server

We wrote about this in March, when WordPress 7.0 still looked imminent: PressBot ships its own MCP server today. A single JSON-RPC 2.0 endpoint at POST /pressbot/v1/mcp, authenticated with WordPress Application Passwords, exposing 73 admin tools — posts, categories, plugins, users, WooCommerce, Shield, analytics, knowledge base, the lot.

It’s not a preview of what 7.0 is going to bring. It’s the same protocol, the same transport, the same agent clients: Claude Code, Cursor, Claude Desktop, Windsurf, whatever comes next. The free tier exposes read-only tools. Pro unlocks writes and the full registry.

When 7.0 eventually ships, your agent workflows don’t break. You’ve been running against MCP the whole time. The core team is just catching up to the interface you’ve already internalized.

What to actually do between now and late May

1. Turn on MCP in PressBot. Settings → MCP. Flip the toggle, generate a WordPress Application Password, and paste the endpoint into your agent client of choice.
2. Start read-only. Even on the free tier, get_posts, get_plugins, get_site_info, and knowledge base search are enough to make agent-assisted editorial reviews and audits feel like a different job.
3. Build your Knowledge Base now. Entries you create today feed both the dashboard agent’s system prompt and MCP resources. When Abilities API lands, your knowledge layer already exists — you won’t be starting from zero on release day.
4. Think in abilities. The mental model of “this site exposes tools the agent can call” is the same mental model 7.0 will demand. Practicing it against PressBot’s tool registry is practice against the future.

The delay isn’t a crisis. It’s a hint.

The fact that WordPress 7.0’s AI infrastructure is ready-but-blocked tells you something about where the platform is going: AI is core now, not a plugin category. The MCP Adapter isn’t a side feature sharing a changelog with RTC — it’s the thing that will make WordPress a first-class citizen in every agent ecosystem being built right now.

Six weeks is a long time to leave that on the shelf. If you’ve been on the fence about wiring an agent into your site, this is the cleanest window you’ll get before the rush.