Automattic, the commercial company behind WordPress, appears eager to stay on trend and has released in GitHub a Model Context Protocol (MCP) plugin and some other offerings. Model Context Protocol is an innovative solution that changes how AI models and applications can interact with WordPress sites, and allows them to transform into intelligent, accessible data and functionality hubs.
The Model Context Protocol (MCP) is an open protocol released in late 2024 by Anthropic designed to standardize how artificial intelligence applications and large language models (LLMs) interact with external data sources and tools.. It moves beyond simple API calls, providing a rich context that allows AI to perform complex tasks, understand relationships between different pieces of information, and even anticipate user needs.
Within the MCP framework, host applications (such as Claude Desktop or Integrated Development Environments) utilize MCP clients to interact with servers. Each client maintains a distinct, stateful session with a server, handling protocol negotiation and message routing. Servers, like the WordPress MCP plugin, provide specific capabilities and expose resources, tools, and prompts. In this ecosystem, Large Language Model (LLM) applications typically function as or drive the host applications, initiating connections, managing client permissions, and coordinating the overall AI/LLM integration with external services. This allows for diverse interactions, from a simple chatbox on a website to sophisticated LLM-driven desktop applications. The WordPress MCP plugin is an implementation of this protocol, specifically tailored for the WordPress ecosystem.
Type “Create 10 draft blog posts titled ‘Create a new News piece post with dummy content.” and hit enter. The AI will immediately call the MCP createPost tool (or equivalent) in a loop to produce those entries. This is a huge time-saver when you need filler content for theme development or want to bootstrap a new site section. Similarly, the agent could update or delete posts in bulk via natural language commands. Other things such as “Install and activate the Gutenberg, Query Monitor, and FakerPress plugins.” can be done as well. Instead of manually downloading or using WP-CLI, an AI can trigger the plugin installation tools provided by the MCP plugin. This could extend to theme activation, toggling settings, or other configuration changes – all by asking the AI, which then calls the proper WordPress REST API endpoints behind the scenes. All these things are possible thanks to MCP and some code by Automattic and other parties, although it requires installation and configuration first.
The WordPress MCP plugin by Automattic is designed with a dual-transport architecture. For established setups, the STDIO Transport provides a WordPress-style REST API endpoint at /wp/v2/wpmcp. This traditional method supports both JWT (JSON Web Token) and Application Password authentication. For modern AI clients seeking direct and low-latency communication, the Streamable Transport offers a JSON-RPC 2.0 compliant endpoint at /wp/v2/wpmcp/streamable. This direct connection relies solely on JWT authentication and supports advanced JSON-RPC 2.0 features such as batch request handling (sending multiple requests in a single call) and notification handling (requests without an expected response).
Its security measures allow users to generate secure, temporary tokens with customizable intervals (from 1 to 24 hours, with a maximum of 30 days configurable) directly from the WordPress Admin interface. These tokens inherit the generating user’s permissions. The plugin also provides granular CRUD Operation Controls, enabling administrators to individually enable or disable “create,” “update,” and “delete” (CRUD) functionalities via MCP tools.
This plugin allows for new possibilities for managing and interacting with their site, particularly for e-commerce operations. Beyond general posts and users, the plugin provides specialized tools for media management (listing, getting, uploading, updating, and deleting media files ), page management (searching, getting, adding, updating, and deleting pages), and even custom post types (listing, searching, getting, adding, updating, and deleting custom post type entries). For e-commerce, WooCommerce tools extend to managing products (searching, getting, adding, updating, deleting products, categories, tags, and brands) and comprehensive sales reports (coupons, customers, orders, products, reviews, and general sales totals).
A barebones WooCommerce installation provides the fundamental tools for online selling and a manual administrative interface, and this plugin as is out of the box adds an AI-accessible layer on top of it. This means an end-user can leverage AI to perform various tasks that would otherwise require navigating through multiple menus and screens. For instance, instead of manually checking product stock or generating sales reports, a user could simply ask an AI: “List all my products,” “Create a new product called ‘Summer T-Shirt’ with a price of $25,” “Show me all orders from last week that are still processing,” or “Summarize my sales for the last 30 days.”.
The plugin offers an Admin Interface with a React-based token management and settings dashboard. This user-friendly interface allows for easy configuration of the plugin’s features. From here, users can enable or disable the main MCP functionality, toggle specific tool types (create, update, delete) for fine-grained control, and even activate Experimental REST API CRUD Tools. This experimental feature provides generic tools capable of interacting with virtually any WordPress REST API endpoint, including those exposed by other plugins, even if they don’t have direct pre-built integrations with MCP. This means an AI could potentially discover, inspect, and execute functions from third-party plugins, provided they expose their functionalities via the WordPress REST API.
Furthermore, the plugin boasts an extensible architecture, supporting custom tools, resources, and prompts. This allows developers to build tailored AI integrations for specific needs, including creating custom PHP classes to directly interact with WooCommerce Subscriptions or other plugins that lack built-in MCP support.
The extensibility of the solution allows for custom developments build on top of this ecosystem. One niche integration is the Elementor MCP server by Aguaitech, which lets AI manipulate Elementor-built pages. It is not build on top of the Automattic plugin, but is another example of what can be done with WordPress and MCP. This server enables create/update/delete operations on Elementor sections and widgets via WordPress REST API calls.
For developers working with landing pages or site designs, it means you could prompt an AI to “rearrange this page’s layout” or “insert a call-to-action banner,” and the MCP server will try to translate that into the appropriate Elementor structure changes. The Elementor MCP server runs as a Node service (invoked with npx) and uses WordPress application passwords for auth, similar to other external servers.
In addition to the main WordPress MCP plugin, Automattic has also released a specialized MCP Server for Gravatar, known as mcp-server-gravatar. This standalone server is designed to provide AI models with access to Gravatar’s functionalities, including user avatars, profiles, and experimental AI-inferred interests. The Gravatar MCP Server is open-source and available on GitHub, and can be easily installed and run using npx with Node.js version 20.0.0 or higher.
Gravatar, short for “Globally Recognized Avatar,” is a service that allows users to create and manage a single, persistent avatar (profile picture) that follows them from website to website. By default, although this is often deactivated by plugins, themes or customizations, when a user comments on a WordPress blog or any other Gravatar-enabled site, their Gravatar image automatically appears next to their name, provided they use the same email address linked to their Gravatar account. While Gravatar is a separate service, it has a deep, long-standing relationship with WordPress due to both being owned by Automattic.
The plugin has features to retrieve comprehensive Gravatar profile information using either a profile identifier (SHA256 or MD5 hash of an email, or a URL slug) or an email address respectively, fetching AI-inferred interests for a Gravatar profile using either a profile identifier or an email address, retrieve avatar images based on an avatar identifier (SHA256 or MD5 hash of an email) or an email address, with options for size, default images, and content rating.
Regarding interest discovering, the entire structure (tool definitions, input/output schemas, and handlers) the plugin uses to do so adheres to the Model Context Protocol. This standardization allows any MCP-compliant AI application to discover, understand, and invoke these Gravatar functionalities without needing prior knowledge of Gravatar’s specific APIs. The AI client simply sends an CallToolRequest with the tool name and arguments, and the MCP server handles the underlying API calls and returns a structured response.
In the Model Context Protocol (MCP) context, a CallToolRequest is a message sent by clients to servers to invoke a specific tool. This request is part of the tools capability that servers can declare during initialization. When a CallToolRequest is sent, the server executes the requested operation and then returns the results of that operation to the client.
By defining clear description fields and inputSchema/outputSchema, the plugin provides rich context to the AI. This allows the AI not just to call an API, but to understand what kind of information it needs to provide, what type of data it will receive, and the purpose of the data.
The fact that mcp-server-gravatar is a separate MCP server demonstrates the protocol’s extensibility. It allows developers to create specialized integrations for various services (like Gravatar, but it serves as example to create custom solutions as well) that can then be exposed to AI models through a common, open protocol, rather than requiring each AI to have bespoke integrations for every service.
Building upon the foundational understanding of the Model Context Protocol (MCP) and its specialized applications like the mcp-server-gravatar, Automattic’s wordpress-reader-mcp is another implementation they offer, particularly integrated into Claude Desktop. It is a WordPress.com Reader API integration for Claude Desktop via the Model Context Protocol (MCP). It allows Claude Desktop to access various WordPress.com functionalities, including reading user feeds, checking notifications, managing tags and followed blogs, and interacting with blog content. It acts as a bridge between Claude Desktop and WordPress.com, handling authentication and API calls.
The code exposes an extensive set of 27 distinct tools, enabling Claude Desktop to manage diverse WordPress Reader functionalities through natural language commands. These tools span critical areas such as feed management, allowing users to retrieve posts from their following feed, access detailed feed information by URL or ID, and so on.
Blog interaction capabilities are multiple, encompassing the ability to follow and unfollow blogs by URL, and to retrieve personalized blog recommendations. The plugin also provides comprehensive tag management, and enables users to get a list of followed tags, subscribe to new ones, unsubscribe from existing tags, and filter posts by specific tags. For content access, it allows the retrieval of individual reader posts, offering control over whether to include the full content to manage response sizes, and provides tools to fetch Automattic (A8C) company posts.
The Comments API integration is particularly comprehensive. Users can retrieve a list of comment notifications, filter them by read status or type (e.g., comments, likes, follows), and mark them as seen or read. Furthermore, it supports updating and permanently deleting comments.
The piece of MCP-related code released by Automattic so far to comment extensively isAutomattic/mcp-wordpress-remote, although is is not a plugin, but something that can be used in combination with one. The mcp-wordpress-remote package from Automattic serves as a proxy server, bridging local Model Context Protocol (MCP) clients with remote WordPress MCP servers. Its primary purpose is to enable AI models and applications to interact with WordPress sites by transforming them into accessible hubs for data and functionality.
It functions as a proxy that facilitates communication between MCP clients and WordPress. Unlike traditional methods that might keep open connections, this proxy makes simple REST API requests to the WordPress website only when necessary, aiming for improved efficiency. It supports both JWT (JSON Web Token) and WordPress Application Password authentication for its WordPress API requests.
The proxy is designed to handle various MCP client requests by forwarding them to the WordPress API. These capabilities include listing and calling tools, listing and reading resources, listing resource templates, subscribing to and unsubscribing from resource updates, listing and getting prompts, setting logging levels, handling completion requests, and listing roots. For example, when a client sends a ListToolsRequest or a CallToolRequest, the proxy processes it and makes the appropriate call to the WordPress API. It also includes a logging mechanism that records incoming requests and outgoing responses for debugging purposes.
In the Model Context Protocol (MCP) context, a ListToolsRequest is a message sent by clients to servers to discover available tools. This request is part of the tools capability that servers can declare during initialization.
With these utilities, no longer are we confined to manual navigation and data entry for every task. After careful implementation we can now tell AI to “list all products with low stock” or “create a new blog post about the latest plugin updates,” and it happens instantly.
The implications are vast, particularly for e-commerce. Managing products, processing orders, and generating sales reports — tasks that once consumed valuable time — can now be asked to an AI through natural language. Beyond WooCommerce, the plugins capabilities extend to media, pages, and even custom post types, all of these out of the box, and the extensible architecture further opens doors for developers to create bespoke AI integrations.
If you wish to contact Álvaro Martínez Majado for professional matters or regarding this post, please email at [email protected].