WordPress in the Age of AI Agents: Abilities API, AI Client and MCP

Summarise this article using AI:

TL;DR
WordPress is preparing to allow AI agents to operate within the CMS, not just users and developers. The Abilities API (since version 6.9) provides a structured way to register what a WordPress website can do; AI Client (since version 7.0) gives plugins and developers a common way to use AI models without having to integrate with each provider separately; and the MCP Adapter translates those capabilities into tools that an external agent can discover and use. Together, these three projects pave the way for a WordPress CMS that can be operated by AI.

In the age of AI agents, where the web is increasingly shaped by artificial intelligence, WordPress has been working on a series of projects focused on two main objectives: building a shared infrastructure that allows developers, plugins and other services to work with AI models in a standardised way, and enabling agents to understand what a WordPress website is capable of doing and interact with it.

This initiative (AI Building Blocks for WordPress) consists primarily of the Abilities API, AI Client and MCP Adapter projects, and marks the beginning of a new landscape in which graphical interfaces are no longer the only way to interact with WordPress.

In this article, I take a closer look at each of these projects to understand why they matter in this new era of WordPress development:

Abilities API: What WordPress Can Do

This is a new API, available since WordPress 6.9, that allows different actors (plugins, themes, agents, etc.) to understand and interact with the functionalities available on a website.

For many years, WordPress has provided ways to interact with these functionalities through PHP functions, hooks, the REST API, etc. However, there has never been a common registry. That is exactly what the Abilities API provides. Through what are known as abilities, a system can understand, in a structured way, what a WordPress website can do.

The WordPress developer documentation describes an ability as a specific unit of functionality registered under a unique name following the namespace/ability-name pattern. An ability has a name and a description, can define the parameters it accepts and the result it returns using JSON Schema, and can include a function that controls who has permission to execute it.

For example, abilities could look something like this:

  • core/read-settings
  • core/read-content
  • core/read-users
  • my-plugin/generate-report

At first glance, they might seem like PHP functions with different names, but there are some important differences. For example, abilities are described in a way that allows other applications to discover and interpret them. This matters because the Abilities API is not strictly an AI API. An ability can be used from PHP, WordPress interfaces, automations, external applications or any other system. What makes it relevant to AI is that this layer turns CMS functions into structured, machine-readable operations. And this is exactly the kind of information an agent needs to understand which tools are available to it.

WordPress also provides REST endpoints for discovering and executing abilities. By default, an ability is not available through REST, so it must be explicitly enabled, and its execution remains restricted by authentication and the permission_callback defined for that ability.

This is particularly important when it comes to agents. Allowing AI to understand WordPress capabilities does not mean giving it unrestricted access to our CMS.

AI Client: A Common Interface for Using AI Models

We have already seen how the Abilities API helps systems discover what WordPress can do. AI Client works in the opposite direction. This interface arrived with WordPress 7.0 and allows WordPress to use artificial intelligence models.

In simple terms, AI Client provides a provider-agnostic PHP API that allows plugins and developers to send prompts to AI models without having to build an entire integration for a specific provider.

Until now, every plugin that wanted to interact with an AI model had to handle authentication, errors, configuration, compatibility, etc. independently. With AI Client, however, we have an intermediate layer that takes care of this, centralising part of the infrastructure and standardising the interaction. It would look something like this:

Before: Plugin 1 -> OpenAI API | Plugin 2 -> Anthropic API | Etc.

Now: Any plugin -> AI Client -> available provider

According to the official documentation, the aim is to reduce duplication of SDKs and configuration systems, and provide a stable infrastructure for transport, credentials, invocation and access controls.

One interesting point mentioned in the documentation is that, in WordPress, this solution is organised into a two-layer architecture:

  1. The AI Client we are already familiar with, an SDK that manages communication with providers, model selection and response normalisation.
  2. A WordPress wrapper that adapts the SDK to WordPress conventions and its ecosystem.

I also think it is important to mention that WordPress does not send information from our website to OpenAI, Anthropic or any other provider simply because AI Client is included in its Core. AI functionality is not integrated directly into the CMS. Instead, WordPress provides the infrastructure on which these integrations can be built.

So far, three official implementations have been developed to integrate the most popular providers: Anthropic, Google and OpenAI. However, anyone can develop their own plugins to integrate other providers without any problems.

Two Different Directions

So far, the overall structure is fairly clear. With AI Client, WordPress can use an AI model, while the Abilities API allows it to describe its own capabilities in a structured way.

What remains is to understand how an external AI agent can discover those capabilities and use them as tools.

This is where the well-known MCP comes in.

MCP: WordPress Speaks the Language of AI Agents

Model Context Protocol. Nothing we are not already familiar with. An open protocol designed to standardise how AI applications connect to other tools or sources of information.

Essentially, it allows an application to discover which tools another system provides and how to use them.

WordPress is working on an MCP Adapter that connects this protocol to the Abilities API, translating WordPress abilities into concepts that an MCP client can understand and exposing them as tools or prompts available to agents.

With this, we now have a much clearer picture:

WordPress registers its capabilities through the Abilities API, the MCP Adapter exposes them using a protocol that agents can understand, and the agent discovers these capabilities (tools) and decides which ones to use to complete a specific task.

With this workflow, WordPress can therefore become a system that can be operated by AI agents.

The New WordPress Paradigm

The key is to understand that the Abilities API, AI Client and MCP are not simply three new WordPress features, but together form a new framework through which AI can interact with the CMS.

AI Client allows WordPress to use AI models, the Abilities API provides a structured way to describe what WordPress can do, and MCP allows agents to discover those capabilities and use them as tools. This is the overall structure we need to keep in mind.

For many years, WordPress has been designed for two types of users: regular users, who managed content through the WordPress admin area, and developers, who worked with code or APIs. Now, the door is opening to a third type of user: AI agents. We will be keeping an eye on how the web continues to evolve.

In this article

Do you find this interesting?

Add VILAX as a preferred source on Google and find our case studies, analyses and content on digital marketing more easily.

Frequently Asked Questions

WORDPRESS FOR AI AGENTS

Your Website, Ready to Communicate with AI Agents

At VILAX, we help you prepare your website for this new era: from structuring your content and functionalities so that AI agents can understand them, to implementing the necessary integrations with the CMS’s new standards. We adapt your website so it does not get left behind when AI agents start interacting with it.