Loading Now

How to Connect Claude or ChatGPT to WordPress With MCP (Step by Step)

Imagine instructing Claude by saying, “Please compose a 500-word article regarding the summer sale and save it as a draft on the site.” Wouldn’t it be convenient if that actually happened? Instead of copying from a chat window and pasting into the editor, the post could directly appear in your WordPress dashboard, awaiting your review.

That’s the functionality made available through MCP. Short for Model Context Protocol, this open standard allows AI assistants like Claude, ChatGPT, or Cursor to interact directly with the systems housing your data. Your WordPress site is one of those systems.

This guide provides an easy-to-follow breakdown of what MCP entails, how linking an assistant to your site can enhance its functionality, and steps to set it all up with a free plugin. We’ll also dedicate time to discussing safety, since granting an AI write access to a live website is not something to be taken lightly.

Let’s commence with the fundamentals.

What Is MCP, Explained Simply?

Model Context Protocol is an open standard that Anthropic made available on November 25, 2024, designed “to connect AI assistants to the systems where data resides.” It was developed at Anthropic by David Soria Parra and Justin Spahr-Summers and has gained traction with OpenAI and Google DeepMind. This is significant because it means one setup is compatible across various AI clients, alleviating the need for individual integrations with each assistant.

Consider MCP as a universal language facilitating communication between two entities. On one end, your site operates a small MCP server that highlights WordPress actions as clearly defined “tools”: such as creating a post, moderating a comment, or reading an SEO title. Each “tool” has a set name, inputs, and outputs, enabling the assistant to understand exactly what it can request and what response to expect. Conversely, the AI client (Claude, ChatGPT, Cursor, etc.) invokes these tools using the protocol when you command it to perform a task.

The outcome? The assistant evolves from being a simple text box into a capable entity that can read from and operate on your site, subject to the boundaries you establish.

mcp-wordpress-diagram-1 How to Connect Claude or ChatGPT to WordPress With MCP (Step by Step)

What Can an AI Assistant Accomplish on Your Site?

It turns out, the capabilities are quite extensive. Based on the current functions offered by WordPress MCP Servers, a connected assistant can:

  • Create, schedule, and publish posts directly from a chat, meaning a prompt such as “Compose a 500-word blog post about X and save it as a draft” results in an actual draft in wp-admin.
  • Moderate comments: approve, hold, classify as spam, or delete.
  • Upload media and automatically generate alt text for it.
  • Edit SEO titles and meta descriptions for six popular SEO plugins: Yoast, Rank Math, AIOSEO, SEOPress, Slim SEO, and The SEO Framework.
  • Provide insights on your data, such as “Which keywords does my homepage rank for?” by interfacing with Google Search Console and GA4 once connected.
  • Handle WooCommerce products and orders, create user accounts, and modify menus.

All these actions trace back in an audit log, creating a record of the AI’s activities and their timestamps. More on that in the safety section, as it deserves special attention.

One point of clarification before proceeding: MCP pertains to granting an external assistant access to your site. If what you seek are AI functionalities within wp-admin (like content creators, visitor chatbots, or AI image tools), that belongs to a different category discussed in this overview of ChatGPT plugins for WordPress.

Two Ways to Connect: The Official Adapter or a Plugin

As of 2026, there are two viable methods for linking an AI assistant to a WordPress site: via the core standards or using a plugin that simplifies the process for you.

Route 1: The Official MCP Adapter (Standards Route). WordPress 6.9 introduced the Abilities API, a central registry for plugins to declare specific “abilities” that a site can perform. The official MCP Adapter, part of the “AI Building Blocks for WordPress” initiative, transforms these registered abilities into MCP tools and makes them accessible over HTTP at /wp-json/mcp/... (or via STDIO through wp mcp-adapter serve). Remote clients connect using the @automattic/mcp-wordpress-remote proxy, authenticated with WordPress Application Passwords. This core-aligned path is future-proof, but requires WordPress 6.9 or newer and some familiarity with developer tools.

Route 2: Easy MCP AI (Free, Plugin Route). Easy MCP AI is a free plugin available on WordPress.org that turns a site into an MCP server with just a few clicks, no command line needed. It comes with 242 built-in tools from the start: 96 core WordPress tools (including posts, pages, media, users, comments, and menus), 92 tools integrating with plugins (46 specifically for WooCommerce along with ACF, The Events Calendar, BuddyPress, and the mentioned SEO plugins), plus 54 data tools that incorporate GA4, Search Console, Semrush, SE Ranking, DataforSEO, and a complimentary Ahrefs Domain Rating check. It requires only WordPress 6.0 and PHP 7.4, and if running on WordPress 6.9+, it automatically discovers Abilities API registrations, allowing both routes to complement each other rather than compete.

Here’s a brief comparison of the two routes:

Official MCP AdapterEasy MCP AI
TypeStandards route, part of the AI Building Blocks initiativeFree WordPress.org plugin
RequiresWordPress 6.9+, developer toolsWordPress 6.0+, PHP 7.4+
AuthenticationApplication Passwords via remote proxyIntegrated OAuth server or bearer tokens
Best suited forDevelopers preferring the core-aligned solutionWebsite owners wanting a simple setup

The following instructions will utilize Easy MCP AI, as it tends to be the more user-friendly option for most site operators.

Step 1: Install Easy MCP AI

In your WordPress dashboard, navigate to Plugins → Add New, search for “Easy MCP AI,” and click Install followed by Activate. This follows a standard plugin installation process: no accounts to create or external services to sign up for.

It’s important to be aware that this plugin ‘phones home’ upon default installation. The external data connections (GA4, Search Console, Semrush, etc.) are opt-in and remain inactive until you, as an admin, save your credentials for them.

Step 2: Generate a Token with the Minimal Required Permissions

Access to your site is governed by tokens, each with its own set of permissions. In the plugin settings, create a token specific to your AI client and determine what actions it can perform.

A sensible starting point is a read-only token. With just read access, the assistant can evaluate your content, answer queries about your posts, and generate reports without altering any text. You can always generate a second, more powerful token later once the initial setup has established its reliability.

easy-mcp-ai-settings-scaled-1 How to Connect Claude or ChatGPT to WordPress With MCP (Step by Step)

A few reassuring details about the internal workings: tokens are stored using a SHA-256 hash, each tool call adheres to WordPress capability checks (thus a token cannot exceed the role it represents), and every token has its own rate limit (60 requests per minute by default). Additionally, IP whitelisting is available if you wish to enhance security.

PRO TIP: Start with a read-only token and monitor it for a week. You will learn the assistant’s usefulness before granting it the authority to publish anything.

Step 3: Connect Claude Desktop Using OAuth

This is where the configuration becomes quite straightforward. The plugin incorporates a built-in OAuth 2.0/2.1 authorization server (featuring PKCE and refresh-token rotation for the conscientious), and it supports dynamic client registration as per RFC 7591. In practical terms, this means connecting Claude Desktop or Cursor can be accomplished through a one-click authorization flow rather than extensive manual configuration: direct the client to your site, approve the request in the browser, and the token exchange occurs automatically.

For clients that do not support OAuth, the bearer token from Step 2 can be utilized as the credential. In either scenario, the assistant will only see the permissions assigned to that specific connection.

After establishing the connection, start with a simple and low-risk query: “List my five most recent posts and their categories.” If the response is accurate, then the pipeline is functioning correctly.

Prioritising Safety: Write Access Holds Real Power

This section is not a mere formality; it’s important to express clearly: an AI assistant granted write access to a live site possesses actual power. It can create, modify, and erase content rapidly, which implies that errors can escalate just as swiftly as productivity. Consider permission choices with the same caution you would apply if granting a new employee admin access from day one.

The positive news is that the available tools provide substantial safeguards. Make sure to utilize all of them:

  • Start with Read-Only. Per-token permissions allow you to begin with an assistant that can observe but not alter.
  • Capability Checks on Each Call. Each tool invocation undergoes validation against WordPress user capabilities, ensuring tokens do not exceed the associated role.
  • Rate Limits. Individual token throttling (60 requests per minute by default) prevents a runaway loop from overloading the site.
  • Audit Trail. Every action taken by the AI is recorded, making it easy to determine “what changed and when.”
  • Change History. Snapshot records of MCP modifications, with a retention period of 90 days. This feature is off by default, so activate it prior to your first write session.
  • Force Draft. An option that keeps AI-generated content in draft status, ensuring anything the assistant composes cannot go live without human consent.
  • Backups. Before enabling any write access, ensure you have recent, recoverable backups of your entire site. This is the precaution you hope never to need.

PRO TIP: Enable Change History and the force-draft option before your first write session. Drafts are easily reviewed, while published errors are much harder to rectify.

What MCP Cannot Accomplish

Let’s address some realities, as the excitement around this technology can be overwhelming:

  • It Cannot Assess Quality. The assistant can draft content in a matter of moments, but judging whether that content is accurate, aligns with your brand, and merits publication is ultimately a human task. It will not replace your editor, and treating it as such is how low-quality material gets published.
  • It Can Only Utilize Granted Tools. If a tool does not expose an action, the assistant cannot execute it, regardless of how nicely you inquire. MCP expands what an AI can access, not invent.
  • It Doesn’t Include Your Data Automatically. Connections to GA4, Search Console, Semrush, and alike are opt-in and require admin-saved credentials. It does not establish connections on its own.
  • It Is Not a Strategic Solution. While the assistant can inform you of your homepage’s keyword rankings, determining the next steps is still your responsibility.
  • The Standards Route Has Minimum Requirements. To follow the official adapter path, your site must be running WordPress 6.9 or later.

Conclusion

Though MCP may sound technical, it is beneficial: a common language that enables the AI assistant you already use to read from and act upon your WordPress site. Here are the main takeaways:

  • MCP is an open standard from Anthropic that is widely adopted across major AI clients, allowing you to use one setup across platforms.
  • A connected assistant can draft posts, moderate comments, update SEO metadata, and respond to queries based on your Search Console and analytics data.
  • Setup comprises three steps: install the plugin, generate a token (starting with read-only access), and connect via OAuth.
  • The safety framework is a core feature: per-token permissions, rate limits, audit logs, forced drafts, and ensuring backups before making any significant changes.

Begin with read-only access, identify where the assistant can genuinely save you time, and only grant more permissions when trust is established. Your future self, the one reviewing a clear audit log instead of untangling a chaos, will appreciate your caution.

Have you linked an AI assistant to your WordPress site yet? What tasks would you feel comfortable delegating to it first? Share your thoughts in the comments section below!

Yay! You made it to the end of the article!

Share this content:


Discover more from Qureshi

Subscribe to get the latest posts sent to your email.

Discover more from Qureshi

Subscribe now to keep reading and get access to the full archive.

Continue reading