Elementor opens its editor to AI agents: Claude and Codex can edit native elements via MCP
Elementor MCP lets clients such as Claude and Codex work with the page's native structure instead of generating throwaway HTML. It is still beta, but it points to where WordPress is heading.
Elementor has launched the public beta of Elementor MCP. In practice, that means a compatible AI agent can connect to WordPress and work with the site’s actual structure: read elements, create layouts, and edit them in a form you can still open and continue working on normally in Elementor.
The name sounds like a minor technical detail, but it changes the workflow in an interesting way. AI no longer has to generate a chunk of HTML for someone to paste in and clean up afterward. It can be given a defined set of allowed operations and use them much like any other application would use an API.
What does Elementor MCP actually do?
MCP, or Model Context Protocol, is a standard that lets AI models discover available tools and call them in a structured way. Elementor exposes page-editing functions through MCP, and clients such as Claude Code, Claude Desktop, Codex, or Cursor can use them.
Explore our services.
According to Elementor, an agent can build Atomic Pages and layouts, edit existing elements, and work with the site’s design system. The result is meant to be a native, editable Elementor structure – not a picture of the page or code generated as a one-off.
Why this matters more than another website generator
Website generators usually solve the first-draft problem. MCP is trying to solve a different one: how do you let an agent perform specific operations inside an existing system without making it pretend to be a human clicking through the interface?
It is the difference between “write me a hero section” and “build a hero section from these components, use the existing colour and typography variables, save it as a draft, then change the button in three places”. The second version is much closer to actual website automation.
WordPress laid the groundwork first
Elementor is not building this in isolation. Since version 6.9, WordPress has had the Abilities API, a standard way to describe functions that the system, a plugin, or an external tool can discover and execute. The official WordPress MCP Adapter can expose those abilities to agents as MCP tools.
That matters because a shared language is starting to emerge between WordPress, plugins, and agents. If this model catches on, integrating AI into a website will not have to mean building a separate custom bot for every plugin.
Beta is still beta
Elementor MCP is currently in beta. Elementor also lists Atomic Editor as a requirement and notes limitations in the current feature set. This is not a sign that you should give an agent full access to a production store tomorrow and tell it to “take care of everything”.
It is, however, worth testing this model in a staging environment. The more interesting question is not even “can AI make a good-looking website?” but “once the work is done, are you left with a normal, readable structure that a human can continue to maintain?”
Where could this actually save time?
- creating many similar sections and subpages from an existing design system,
- making bulk changes to repeated elements,
- building the first version of a layout from a prepared brief,
- organising a site around defined components and variables,
- automating tasks that currently require a lot of repetitive clicking.
WebsiteGarage’s take
The most interesting part of this implementation has nothing to do with the usual promise that “AI will build your website for you”. What matters is that we are finally getting a sensible layer between the model and the CMS.
If an agent is given a small, clearly defined set of tools and works with native editor structures, it is easier to limit its permissions, inspect the result, and roll changes back. That is a much healthier direction than giving a model browser access and hoping it clicks the right “Save” button.
At the same time, MCP does not solve the quality problem. An agent can still execute a bad plan very efficiently. The design system, content structure, accessibility, and a decent brief still have to be there. Automation does not fix chaos – it just lets you reproduce it faster.
Discover more articles
- An AI agent was only supposed to book a gym class. Along the way, it started hacking the system
- Website redesign or rebuild? When to fix it - and when to stop keeping a wreck alive
- WordPress maintenance: what a professional service should actually include
- The Cyber Resilience Act reporting portal is live. For manufacturers, the 24-hour clock starts when they become aware of the problem
- Researchers found over 6,800 AI chat links in DNS data. No hacking required
- Admin Menu Editor Pro distributed a web shell through its official update channel. The replacement version was compromised too
- AI can recommend a brand without sending traffic. New study shows the difference between a mention, citation and click
- When a placeholder domain becomes a trap: a real-world example
- How NOT to commission a website