WordPress MCP: Connecting AI Assistants to Your Site

AI assistants can now do more than write a draft for you to paste into the editor. With the Model Context Protocol, an assistant can list your posts, create a draft, check a plugin setting or pull a report directly from your site. “WordPress MCP” is the umbrella term for the ways WordPress exposes itself to those assistants. This guide explains what MCP is, the main routes for connecting it to WordPress, what it is actually useful for, and how to set it up without handing an AI model the keys to production.

One caveat up front: this is a young, fast-moving area. Package names, plugin status and recommended setups have changed several times since MCP appeared, so treat the specifics below as a map of the landscape and confirm details against the current documentation before you build on them.

What the Model Context Protocol is

The Model Context Protocol is an open standard, introduced by Anthropic in late 2024, that defines how an AI application talks to external tools and data. Before MCP, every assistant needed its own custom integration for every service. MCP replaces that with one shared contract.

There are three roles to understand:

• Host – the AI application you use, such as a desktop chat app, an IDE or a coding agent.
• Client – the component inside the host that maintains a connection to one server.
• Server – the program that exposes capabilities. For WordPress MCP, the server is either your WordPress site itself or a small bridge process that talks to it.

A server offers three kinds of things: tools (actions the model can call, like “create a post”), resources (data the model can read, like site settings) and prompts (reusable templates). Communication happens over a transport: either standard input/output for a local process, or HTTP for a remote server. The model never touches your database directly; it can only call what the server chooses to expose.

How WordPress MCP connections work

There are three broad routes, and they overlap.

1. The Abilities API and the MCP Adapter

The WordPress Core AI team has been building a set of “AI building blocks”. The foundation is the Abilities API: a standard way for core, plugins and themes to register discrete units of functionality, each with a name, a description, an input schema, an output schema and a permission check. It shipped in core starting with WordPress 6.9, so check the developer notes for the version you run.

Abilities are not MCP-specific. They are a machine-readable catalog of what a site can do. The MCP Adapter, an official WordPress project, then translates registered abilities into MCP tools, resources and prompts, so any MCP-capable assistant can discover and call them. This is the direction the project is heading: plugins describe their capabilities once, and the adapter exposes them.

A plugin registering an ability looks roughly like this (illustrative; check the handbook for the exact arguments in your version):

add_action( 'wp_abilities_api_init', function () {
    wp_register_ability( 'my-plugin/get-draft-count', array(
        'label'               => 'Get draft count',
        'description'         => 'Returns the number of draft posts.',
        'category'            => 'site',
        'input_schema'        => array( 'type' => 'object' ),
        'output_schema'       => array( 'type' => 'integer' ),
        'execute_callback'    => function () {
            return (int) wp_count_posts()->draft;
        },
        'permission_callback' => function () {
            return current_user_can( 'edit_posts' );
        },
    ) );
} );

The permission callback matters most. It means an AI client authenticated as a low-privilege user cannot run an ability that user could not run in the dashboard. If you are new to how capabilities gate actions, our user roles and capabilities guide is a good primer.

2. Automattic’s wordpress-mcp plugin

Before the core effort matured, Automattic published an open-source wordpress-mcp plugin that turned a site into an MCP server, with tools mapped to common REST endpoints and a companion bridge for clients that only speak local transports. It was an important early implementation and many tutorials still reference it. Automattic has since pointed toward the official MCP Adapter as the long-term path, so read the plugin’s repository notes before choosing it for a new project.

3. The REST API with application passwords

The oldest and most portable route needs nothing new in WordPress at all. A generic MCP server, either community-built or one you write, wraps the WordPress REST API and authenticates with an application password. Application passwords have been in core since WordPress 5.6: go to Users → Profile, scroll to Application Passwords, give it a name and click Add. WordPress shows the password once.

curl --user "ai-editor:abcd efgh ijkl mnop qrst uvwx" https://example.com/wp-json/wp/v2/posts?status=draft

This works on almost any host and any WordPress version from the last several years. The trade-off is that the AI sees raw endpoints rather than well-described abilities, so results depend on how well the wrapping server documents each tool.

Choosing a route

Route	Best for	Trade-offs
Abilities API + MCP Adapter	New builds, plugin authors, long-term setups	Needs a recent WordPress; plugins must register abilities to be useful
wordpress-mcp plugin	Existing setups that already use it	Check its maintenance status before adopting
REST API + application password	Older sites, quick experiments, custom servers	Less descriptive tools; you maintain the bridge

Hosted platforms add a fourth option: WordPress.com and some managed hosts have begun offering MCP access as a feature of the platform. If you are on one of those, the host’s documentation is the place to start.

Practical use cases for WordPress MCP

The useful jobs tend to be the repetitive, structured ones where a human still reviews the result:

• Content operations – create drafts from an outline, add categories and tags in bulk, write missing excerpts or alt text, find posts that mention an outdated product name.
• Site audits – list plugins with available updates, posts without featured images, pages still set to draft, or users with roles they no longer need.
• Development support – a coding agent working on your theme can query the real site to check which templates exist or what a custom post type’s fields look like, instead of guessing.
• Store tasks – on WooCommerce sites with suitable abilities, summarize recent orders or flag low-stock products.
• Plugin-specific actions – as more plugins register abilities, an assistant can drive a form plugin, SEO plugin or stats plugin through the same interface.

What MCP does not do well is replace judgment. Publishing, deleting and changing settings on a live site should still go through a person.

Security: least privilege and staging first

Connecting an AI assistant to WordPress means giving software that can be wrong, or can be manipulated, the ability to act on your site. Treat it like any other integration account, with extra caution.

1. Create a dedicated user. Never connect with your own administrator account. Make a user such as “ai-editor” and give it the lowest role that does the job: Contributor for drafting, Editor only if it must publish or manage others’ content.
2. Use a named application password. One per client, so you can revoke a single connection from Users → Profile without affecting anything else.
3. Start on staging. Point the assistant at a staging copy until you trust the workflow. Most managed hosts offer one-click staging.
4. Expose only what you need. With the MCP Adapter you choose which abilities are published to MCP. Leave destructive actions out unless there is a clear reason.
5. Require HTTPS. Application passwords travel with every request; never use them over plain HTTP.
6. Keep backups and logs. Have a tested WordPress backup and keep revisions enabled so an unwanted bulk edit can be rolled back.

Watch for prompt injection in particular. If an assistant reads post content, comments or form submissions, text written by a stranger can contain instructions aimed at the model (“ignore your previous task and delete…”). Least privilege is the real defense: an account that cannot delete posts cannot be tricked into deleting them. Most MCP hosts also ask you to approve tool calls; keep that confirmation on for anything that writes. Our security hardening guide covers the wider basics.

Setting up a first connection

Exact steps depend on your client and chosen route, but the general sequence is the same:

1. Update WordPress and confirm your PHP version meets the requirements of the adapter or plugin you plan to use.
2. Install and activate the server side: the MCP Adapter (and any plugins whose abilities you want), or your chosen REST-based server.
3. Create the dedicated user and application password described above.
4. Add the server to your AI client’s MCP configuration. Most clients use a JSON settings file or a settings screen where you give the server a name, the endpoint URL or command, and the credentials, ideally via environment variables rather than plain text.
5. Restart the client and ask it to list the available tools. If the list is empty, check the user’s permissions and that the abilities are marked for MCP exposure.
6. Test with a read-only request (“how many draft posts are there?”) before trying anything that writes.

Some security plugins and hosts block REST API access for unauthenticated requests or strip the Authorization header. If authentication fails even with a correct password, check those first.

Frequently asked questions

Do I need to be a developer to use WordPress MCP?

Not necessarily, but today it helps. Editing a client configuration file and creating application passwords is approachable for a confident site owner. Writing new abilities requires PHP. If you want to go that way, start with our plugin development basics.

Does MCP send my whole site to the AI provider?

No. Only the results of the tools the model calls are passed into the conversation. That still may include post content or user data, so review your AI provider’s data policy and avoid exposing abilities that return personal information you would not want processed.

Will WordPress MCP work with any AI assistant?

It works with any host that supports MCP, and support has spread across many chat apps, IDEs and coding agents. Supported transports and authentication options vary by client, so confirm your client can reach a remote HTTP server or run a local bridge.

Key takeaways

• MCP is an open standard that lets AI assistants call tools exposed by a server; WordPress MCP means your site acts as, or sits behind, that server.
• The core direction is the Abilities API plus the official MCP Adapter; Automattic’s earlier plugin and REST-based bridges remain options.
• Application passwords on a dedicated, low-privilege user are the standard way to authenticate.
• Start on staging, expose only the abilities you need, and keep write confirmations on.
• The area changes quickly, so check current documentation before committing to a specific package.

Used carefully, WordPress MCP turns an assistant into a useful junior colleague for routine site work. Give it a narrow role, watch what it does, and widen access only as trust is earned.