Version: October 2025 Based on: official Core AI Team documentation and the workshop "Turn your local WordPress install into your AI coding assistant" (WordCamp US 2025, Jonathan Bossinger)
- https://make.wordpress.org/ai/
- https://jonathanbossenger.com/wcus2025/
- https://www.youtube.com/watch?v=W7moIT9p0NU
WordPress is developing an open infrastructure called the AI Layer, designed to allow any artificial intelligence model or agent (GPT, Claude, Ollama, etc.) to interact with the WordPress ecosystem using interoperable APIs and open standards.
💡 “WordPress isn’t building an AI — it’s building the language that lets any AI speak WordPress.”
This approach is implemented through the AI Building Blocks, a set of foundational components that define the interoperability layer between WordPress and AI models.
| Component | Primary Role | Status | Public Distribution |
|---|---|---|---|
| Abilities API | Defines what “abilities” or actions WordPress can execute via AI. | ✅ Stable | WordPress/abilities-api |
| PHP AI API | Server-side layer that runs AI models via adapters (OpenAI, LocalAI, etc.). | ✅ Active | — |
| MCP Adapter | Turns WordPress into a Model Context Protocol (MCP) server. | ✅ Public | WordPress/mcp-adapter |
The Abilities API allows registering, listing, and executing abilities (AI-powered actions) from PHP or JavaScript. Each ability defines its title, description, input/output schema, and execution callback.
wp_register_ability( 'get_site_info', [
'title' => 'Get Site Info',
'description' => 'Returns basic site metadata',
'outputSchema' => [
'type' => 'object',
'properties' => [
'site_name' => [ 'type' => 'string' ],
'wp_version' => [ 'type' => 'string' ],
],
],
'execute' => function() {
return [
'site_name' => get_bloginfo( 'name' ),
'wp_version' => get_bloginfo( 'version' ),
];
},
] );These abilities can be exposed as MCP tools, allowing external LLMs to discover and use them directly within their development environment.
The PHP AI API is the layer that abstracts communication between WordPress and AI models. It provides adapters for multiple providers and handles authentication, prompts, and contextual management.
use WP_AI\Models\TextModel;
$model = new TextModel( 'gpt-4o' );
$response = $model->generate( 'Summarize this post' );| Method | Usage | Status |
|---|---|---|
| Cookies + Nonce | Block editor authentication | ✅ Active |
| Application Passwords | Local or remote MCP connections | ✅ Active |
| permission_callback() | Fine-grained ability access control | ✅ Active |
| OAuth / JWT Tokens | Advanced permission systems | 🧪 In design |
The MCP Adapter is based on JSON-RPC 2.0, exposing registered abilities as callable methods to AI models.
Example:
{
"jsonrpc": "2.0",
"method": "run",
"params": { "ability": "get_site_info" },
"id": 1
}Response:
{
"jsonrpc": "2.0",
"result": {
"site_name": "Local WP",
"wp_version": "6.8"
},
"id": 1
}The MCP Adapter turns a WordPress site into a Model Context Protocol (MCP) server, enabling communication with AI-powered editors such as VS Code, Claude, or GitHub Copilot.
Architecture Overview:
AI Editor (VS Code / Claude / Copilot)
│ JSON-RPC (MCP)
▼
MCP Remote Bridge (Node.js)
│ HTTP
▼
WordPress MCP Adapter Plugin
│
▼
Abilities API → PHP AI API → Model
🔗 Repository: Automattic/mcp-wordpress-remote 📦 npm: @automattic/mcp-wordpress-remote
A Node.js tool that acts as a bridge between WordPress (MCP server) and AI development environments like VS Code or Claude.
{
"mcpServers": {
"wordpress": {
"command": "npx",
"args": ["-y", "@automattic/mcp-wordpress-remote"],
"env": {
"WP_API_URL": "https://your-wordpress-site.com"
}
}
}
}| Type | Project | URL | Status |
|---|---|---|---|
| 🧩 Plugin | Abilities API | WordPress/abilities-api | ✅ Active |
| 🌐 Plugin | MCP Adapter | WordPress/mcp-adapter | ✅ Active |
| ⚙️ CLI / Bridge | MCP WordPress Remote | Automattic/mcp-wordpress-remote | 🧩 Public |
| 🧩 PHP AI API (in development) | — | — | 🔧 Internal development |
| 🧩 Example | AI Experiments MCP Server | jonathanbossenger/ai-experiments-mcp-server | 🧠 Workshop example |
| 📘 Documentation | Core AI Team Blog | make.wordpress.org/ai | Active |
| Repository | Historical Role | Status |
|---|---|---|
| Automattic/ai-experiments | Original sandbox for Abilities and PHP AI API | Archived |
| Automattic/wordpress-mcp | First MCP Adapter prototype | Replaced |
| Automattic/wp-feature-api | Early experiments on feature APIs | Historical |
📝 Note: The Automattic/wp-ai-server repository mentioned in internal documents never existed publicly; it was a temporary internal name used during early PHP AI API development.
| Milestone | Date | Status |
|---|---|---|
| AI Experiments Plugin (Automattic) | Q1 2025 | ✅ Completed |
| Core AI Team formation | Q2 2025 | ✅ Completed |
| Abilities API (public plugin) | Q3 2025 | ✅ Completed |
| MCP Adapter migrated to WP org | Q3 2025 | ✅ Completed |
| MCP WordPress Remote (CLI) | Q3 2025 | ✅ Completed |
| Experimental Core integration | Q4 2026 | 🧩 Planned |
-
Install and activate the plugins:
git clone https://github.com/WordPress/abilities-api git clone https://github.com/WordPress/mcp-adapter
-
Configure the MCP client in the LLM like for example:
{
"mcpServers": {
"wordpress": {
"command": "npx",
"args": ["-y", "@automattic/mcp-wordpress-remote"],
"env": {
"WP_API_URL": "https://your-wordpress-site.com"
}
}
}
}- Register custom abilities via the PHP or JS API.
- Connect VS Code or Claude to the local MCP server.
- Explore real examples: jonathanbossenger/ai-experiments-mcp-server
WordPress is moving toward an open, modular, and interoperable integration of artificial intelligence, focused on enabling communication between AI models and the CMS through open standards.
The three current technical pillars are:
- Abilities API
- PHP AI API
- MCP Adapter
Together, these tools position WordPress as an “AI-native” platform, ready to interoperate with any AI model or agent using open standards such as MCP.
🧠 “AI in WordPress isn’t a feature — it’s a new interoperability layer for the entire ecosystem.”