Blog · Updated 26 October 2026

What Is MCP? A Plain-English Guide For Website Owners

In December 2025, Anthropic counted more than 10,000 active public MCP servers. Thirteen months before that, the protocol didn’t exist. That kind of growth is why the term now turns up in webinars, plugin changelogs and hosting newsletters, usually with no explanation attached.

Take the owner of a physiotherapy clinic in Singapore. The clinic’s WordPress site takes bookings, lists, say, four therapists and explains what a sports massage costs. Their web designer said “MCP” twice in one call, a plugin update mentioned it, and now they have typed “what is MCP” into Google to find out whether they should care.

This post is the answer we would give them: what the Model Context Protocol is, who backs it, how it works, what it means for a WordPress site and where it stops being useful. For the full picture of AI assistants on WordPress, start with our guide to WordPress MCP.

What Does MCP Stand For, And Who Made It?

MCP stands for Model Context Protocol, an open standard released by Anthropic on 25 November 2024 for connecting AI assistants to the systems where data lives. “Model” is the AI, “context” is the information and actions it can reach, and “protocol” means an agreed set of rules any company can follow.

Before MCP, connecting your booking system to three assistants meant three separate integrations. MCP replaced that with one shape every assistant understands.

The adoption happened fast:

• March 2025 - OpenAI added MCP to its Agents SDK, with ChatGPT support following.
• During 2025 - Google built it into its Gemini developer tools, and Microsoft announced it for Windows at its Build conference in May.
• 9 December 2025 - Anthropic handed MCP to the new Agentic AI Foundation under the Linux Foundation, co-founded with Block and OpenAI and backed by Google, Microsoft, AWS, Cloudflare and Bloomberg.

That last step matters for you. A standard owned by one company can be dropped on a whim; one run by a neutral foundation is a safer bet to still exist in five years. The official MCP documentation lists Claude, ChatGPT, Visual Studio Code and Cursor among the apps that support it today.

The short version: OpenAI, Google and Microsoft all back it, so time spent learning it won’t be wasted.

How Does MCP Work? Client, Server And Tools

MCP works like a USB-C port for AI: the assistant has one socket, and anything built to the standard plugs into it. The official docs use that exact comparison, and it holds up.

There are three parts, and the names sound more technical than they are.

What Is An MCP Server?

An MCP server is a small program that tells an AI assistant, “here are the things I can do, and here’s what I need to do each one.” It runs on your computer or on the internet, and it’s the part that actually touches your website or data.

What Is A Tool?

A tool is a single, named action with a clear description, like “create a post” or “clear the cache”. The assistant reads the list of tools when it connects, decides which one fits your request, and sends the details. The server does the work and reports back.

So the assistant never logs into your site the way you do. It asks the server, and the server decides what is allowed.

Connectors Vs Coding Agents: Two Ways You’ll Meet MCP

You’ll meet MCP in one of two places: as a connector inside a chat app, or as a setting inside a coding tool. They use the same standard, but they feel very different.

Connectors are what most business owners will use. In Claude, you paste a web address and sign in. ChatGPT offers a similar custom MCP option on some paid plans, and which plans can make changes keeps shifting; see how to connect ChatGPT to WordPress. After that, you type requests as normal.

Coding agents are tools like Claude Code, Cursor, Codex and Visual Studio Code with GitHub Copilot. They can use the same online servers, but they can also run MCP servers on your own computer, with far more reach into your files.

Our take: if you don’t write code, a connector in the chat app you already pay for is the whole story. Coding agents are your developer’s business.

What Can MCP Do For A WordPress Site?

With an MCP server attached, an AI assistant can do most of the routine jobs you’d otherwise log into wp-admin for. Which ones depends on the server’s tools, but a typical list looks like this:

1. Draft and publish content - write a blog post about ankle sprains and save it as a draft for you to check.
2. Edit existing pages - change opening hours or a price on the Services page.
3. Report on the site - list installed plugins and flag the ones with updates waiting.
4. Run maintenance - clear the cache after a change, or take a backup before one.
5. Fix small problems - read an error message and suggest, or apply, a fix.
6. Build something new - set up a fresh WordPress site with a theme and starter pages.

WordPress now has the plumbing for this. Version 6.9 introduced the Abilities API, a standard way for WordPress and plugins to describe what they can do. An official MCP Adapter, first announced in July 2025, turns those abilities into MCP tools. It works on any host, but on its own it only exposes what WordPress core and your other plugins register, which today is a short list.

The other route is a server your host runs for you, so there’s nothing to install. Each approach has trade-offs in setup and control, which we compare in WordPress MCP servers compared.

The verdict: for a small site, MCP turns “I’ll do it when I have an hour” jobs into one sentence typed into a chat window.

Walkthrough: The Clinic Changes Its Opening Hours

Here is an illustrative run of what happens when the clinic owner asks an assistant to update the opening hours. Assume they have already connected the site once.

1. The owner types the request. “We now open at 8am on Saturdays instead of 9am. Update the Contact page.”
2. The assistant checks its tools. It sees the site’s server offers “read a page” and “update a page”.
3. It reads first. It calls “read a page” for Contact and finds the line with Saturday hours.
4. It asks permission. Most chat apps show a prompt before a tool that changes something. They click Allow.
5. The server does the work. It saves the new text through WordPress, exactly as if the owner had edited it by hand.
6. The assistant reports back. “Saturday hours now read 8am to 1pm. Here’s the link to check.”

Total time: under a minute. Notice step 4: the assistant didn’t decide on its own to touch the live site. The owner approved it, and the server only allowed what it was built to allow.

What Are The Limits Of MCP?

MCP only carries requests between the assistant and your site. It does nothing to make the assistant more accurate or more trustworthy. Know these limits before you connect anything:

• The AI can still be wrong. It can misread “Saturday” as “Sunday” and save it confidently. Check every live change.
• Safety comes from the server, not the protocol. MCP carries the request. Whether a tool can delete your whole site depends entirely on how that server was built and what you allowed.
• Text can carry hidden instructions. A web page or email the assistant reads can contain instructions aimed at it. This is called prompt injection, and no standard fully solves it yet.
• Local servers run with your permissions. MCP’s own security guidance warns that a server running on your computer can reach whatever you can, so only install ones from sources you trust.
• Quality varies. Anyone can publish a server, and plenty of public ones are hobby projects nobody maintains.

We cover which permissions to give an assistant, and which to keep back, in WordPress and AI assistants: what to let them touch.

One more limit: MCP only helps if something on the other end speaks it. Oxyforge runs WordPress only, so if your clinic site is on Wix or Squarespace, our server is no use to you. Check what your own platform supports first.

Does Your Business Need MCP Right Now?

Most small businesses don’t need MCP today, and nothing breaks if you ignore it for another year. It’s worth trying if you already use Claude or ChatGPT most days and you still make your own website edits.

The case for waiting: your site works without it, and setup steps are still changing. The case for trying: if you spend an hour a week on small site edits, a connected assistant can cut that to minutes, and a test costs you an afternoon.

Our advice: don’t connect your live booking site on day one. Try it on a copy first.

What To Do Next

Start with one small, low-risk test.

• Pick the assistant you already use. Claude and ChatGPT both support MCP connectors.
• Use a test site, not your live one. A staging copy or a throwaway WordPress install is ideal.
• Give it one boring job. “List my plugins” or “draft a post” will show you how it behaves.
• Read every permission prompt before you click Allow.

If you want a test site without setting anything up, Oxyforge runs a public MCP server that needs no account. Ask your assistant to build a WordPress site and it creates a real, password-protected install that stays up for 24 hours unless you claim it. You can see how the Oxyforge MCP server works and pick your assistant from there.