What if you could tell Claude, “Draft a 500-word post about the summer sale and save it as a draft on the site,” and it simply happened? No copying from a chat window, no pasting into the editor. The post shows up in your WordPress dashboard, ready for review.
That is exactly what MCP makes possible. MCP stands for Model Context Protocol, an open standard that lets AI assistants like Claude, ChatGPT, or Cursor talk directly to the systems where your data lives. Your WordPress site is one of those systems.
This guide walks through what MCP is in plain words, what connecting an assistant to your site actually enables, and how to set everything up with a free plugin. It also spends real time on safety, because giving an AI write access to a production website is not something to do casually.
Let’s start with the basics. 👇
What is MCP, in plain words?
Model Context Protocol is an open standard that Anthropic open-sourced on November 25, 2024, built “for connecting AI assistants to the systems where data lives.” It was created at Anthropic by David Soria Parra and Justin Spahr-Summers, and it has since been adopted by OpenAI and Google DeepMind. That last part matters: one setup works across AI clients, instead of a separate integration for every assistant.
Think of MCP as a common language between two sides. On one side, your site runs a small MCP server that exposes WordPress actions as typed “tools”: create a post, moderate a comment, read an SEO title. Typed simply means each tool has a defined name, inputs, and outputs, so the assistant knows exactly what it can ask for and what it will get back. On the other side, an AI client (Claude, ChatGPT, Cursor, and friends) calls those tools over the protocol when you ask it to do something.
The result: the assistant stops being a clever text box and becomes something that can actually read from and act on your site, within limits you choose.
What can an AI assistant actually do on your site?
Quite a lot, it turns out. Based on what current WordPress MCP servers expose, a connected assistant can:
- ✍️ Draft, schedule, and publish posts straight from chat. A prompt like “Write a 500-word blog post about X and save it as a draft” produces a real draft in wp-admin.
- 💬 Moderate comments: approve, hold, mark as spam, or delete.
- 🖼️ Upload media and generate alt text for it automatically.
- 🔍 Edit SEO titles and meta descriptions across six major SEO plugins: Yoast, Rank Math, AIOSEO, SEOPress, Slim SEO, and The SEO Framework.
- 📊 Answer questions from your data, like “What keywords does my homepage rank for?” by querying Google Search Console and GA4 (once you connect those accounts).
- 🛒 Manage WooCommerce products and orders, create users, and edit menus.
Every one of those actions lands in an audit trail, so there is a record of what the AI did and when. More on that in the safety section, because it deserves its own spotlight.
One clarification before moving on: MCP is about giving an outside assistant access to your site. If what you actually want is AI features inside wp-admin (content generators, chatbots for visitors, AI image tools), that is a different category, covered in this roundup of ChatGPT plugins for WordPress.
Two routes in: the official adapter or a plugin
In 2026 there are two practical ways to connect an AI assistant to a WordPress site: the core-standards route, or a plugin that handles the plumbing for you.
Route 1: the official MCP Adapter (the standards path). WordPress 6.9 shipped the Abilities API, a core registry where plugins declare discrete “abilities” a site can perform. The official MCP Adapter, part of the “AI Building Blocks for WordPress” initiative, converts those registered abilities into MCP tools and serves them over HTTP at /wp-json/mcp/... (or over STDIO via wp mcp-adapter serve). Remote clients connect through the @automattic/mcp-wordpress-remote proxy, authenticated with WordPress Application Passwords. This is the core-aligned, future-proof route, but it needs WordPress 6.9 or newer and some comfort with developer tooling.
Route 2: Easy MCP AI (Free, the plugin path). Easy MCP AI is a free WordPress.org plugin that turns a site into an MCP server in a few clicks, no command line required. It exposes 242 tools out of the box: 96 core WordPress tools (posts, pages, media, users, comments, menus, taxonomies, revisions, settings), 92 plugin-integration tools (46 for WooCommerce alone, plus ACF, The Events Calendar, BuddyPress, and the SEO plugins named above), and 54 data tools covering GA4, Search Console, Semrush, SE Ranking, DataforSEO, and a free Ahrefs Domain Rating check. It requires only WordPress 6.0 and PHP 7.4, and on sites running WordPress 6.9+ it also auto-discovers Abilities API registrations, so the two routes complement each other rather than compete.
Here is the short version of how the two routes compare:
| Official MCP Adapter | Easy MCP AI | |
|---|---|---|
| Type | Standards route, part of the AI Building Blocks initiative | Free WordPress.org plugin |
| Requires | WordPress 6.9+, developer tooling | WordPress 6.0+, PHP 7.4+ |
| Auth | Application Passwords via a remote proxy | Built-in OAuth server or bearer tokens |
| Best for | Developers who want the core-aligned path | Site owners who want a click-based setup |
The walkthrough below uses Easy MCP AI, since it is the more approachable of the two for most site owners.
Step 1: Install Easy MCP AI
In your WordPress dashboard, go to Plugins → Add New, search for “Easy MCP AI,” then click Install and Activate. It is a standard plugin install: no account to create, no external service to sign up for.
Worth knowing at this point: nothing phones home on a default install. The external data connections (GA4, Search Console, Semrush, and similar) are opt-in and stay inactive until you save credentials for them yourself as an admin.
Step 2: Create a token with the least power needed
Access to your site is controlled through tokens, and each token gets its own permissions. In the plugin’s settings, create a token for your AI client and choose what it is allowed to do.
The smart starting point is a read-only token. With read access alone, the assistant can inventory your content, answer questions about your posts, and pull reports, all without being able to change a single character. You can always create a second, more powerful token later once the setup has earned your trust.
A few reassuring details about what happens under the hood: tokens are stored SHA-256 hashed, every tool call passes WordPress capability checks (so a token can never do more than the user role behind it), each token has its own rate limit (60 requests per minute by default), and IP whitelisting is available if you want to lock access down further.
💡 PRO TIP: Start with a read-only token and live with it for a week. You will learn what the assistant is useful for before you hand it the keys to publish anything.
Step 3: Connect Claude Desktop with OAuth
Here is where the setup gets pleasantly boring. The plugin runs a built-in OAuth 2.0/2.1 authorization server (with PKCE and refresh-token rotation, for the security-minded), and it supports dynamic client registration under RFC 7591. In practice, that means connecting Claude Desktop or Cursor is a one-click authorization flow instead of an afternoon of hand-editing config files: point the client at your site, approve the request in the browser, and the token exchange happens on its own.
For clients that do not speak OAuth, the bearer token from Step 2 works as the credential instead. Either way, the assistant only ever sees the permissions you granted to that specific connection.
Once connected, try something small and harmless first: “List my five most recent posts and their categories.” If the answer comes back correct, the pipeline works. 🎉
Safety first: write access is real power 🛡️
This section is not a formality, so it gets said plainly: an AI assistant with write access to a production site holds real power. It can create, edit, and delete content at machine speed, which means mistakes scale just as fast as productivity. Treat the permission decisions below the way you would treat giving a new hire admin access on day one.
The good news is that the tooling gives you serious guardrails. Use all of them:
- Read-only first. Per-token permissions let you start with an assistant that can look but not touch.
- Capability checks on every call. Each tool call is validated against WordPress user capabilities, so tokens cannot exceed the role behind them.
- Rate limits. Per-token throttling (60 requests per minute by default) keeps a runaway loop from hammering the site.
- Audit log. Every AI action is recorded, so “what changed and when” is a lookup, not a mystery.
- Change History. Before/after snapshots of MCP writes, with a 90-day retention window. It is off by default, so turn it on before your first write session.
- Force draft. A setting that keeps AI-created content in draft state, so nothing the assistant writes can go live without a human pressing publish.
- Backups. Before enabling any write access, confirm you have recent, restorable backups of the whole site. This is the seatbelt you hope to never need.
⚡ PRO TIP: Enable Change History and the force-draft setting before your first write session. Drafts are easy to review; published mistakes are not.
What MCP cannot do
Honesty section, because the hype around this stuff is loud:
- 🧠 It cannot judge quality. The assistant can draft a post in seconds, but whether that post is accurate, on-brand, and worth publishing is still a human decision. It will not replace your editor, and treating it as if it does is how thin content gets published.
- 🔧 It can only use the tools it is given. If no tool exposes an action, the assistant cannot perform it, no matter how nicely you ask. MCP expands what an AI can reach, not what it can invent.
- 🔌 It does not come with your data attached. GA4, Search Console, Semrush, and similar connections are opt-in and need credentials you save as an admin. Nothing connects itself.
- 📈 It is not a strategy. The assistant can tell you which keywords your homepage ranks for. Deciding what to do about it is still your job.
- 🧩 The standards route has a floor. If you want the official adapter path, your site needs to run WordPress 6.9 or newer.
Wrapping up
MCP sounds like plumbing, and it is, but it is the useful kind: a common language that lets the AI assistant you already use read from and act on your WordPress site. The practical takeaways:
- MCP is an open standard from Anthropic, now adopted across the major AI clients, so one setup travels with you.
- A connected assistant can draft posts, moderate comments, edit SEO meta, and answer questions from your Search Console and analytics data.
- Setup is three steps: install the plugin, create a token (read-only first), connect via OAuth.
- The safety model is the feature: per-token permissions, rate limits, audit logs, force draft, and backups before anything else.
Start read-only, see where the assistant actually saves you time, and widen permissions only when the trust is earned. Your future self, the one reviewing a clean audit log instead of untangling a mess, will be grateful. 👍
Have you connected an AI assistant to your WordPress site yet? What would you trust it with first? Let us know in the comments section below!
FREE GUIDE
4 Essential Steps to Speed Up Your WordPress Website
Follow the simple steps in our 4-part mini series and reduce your loading times by 50-80%. 🚀
