An MCP server for your CMS: what it is, what it changes.

MCP in two sentences

The Model Context Protocol (MCP) is an open protocol describing how an AI assistant discovers and uses external tools. An MCP server advertises a list of tools with their parameters and descriptions; the assistant, the MCP client, calls them when the task requires it.

Put simply: instead of handing you text to paste into your site's admin, the assistant can carry out the operation itself, under the same rules as you.

What an MCP server brings to a CMS

A CMS is ideal ground for MCP: content is structured, versioned and subject to permissions. A well-designed MCP server typically exposes:

• Discovery: summarising a site (languages, layouts, theme), listing component types and reading their schemas, so the agent knows which fields to fill in.
• Reading: browsing pages, objects and the media library, and searching for existing content before creating a duplicate.
• Writing: creating, updating or cloning a page, always as a revision.
• Translation: creating a page's variant in another language and translating its text.
• Media: uploading an image to the media library, or reading a supplied document to turn it into a page.

In Plumixia, these tools have explicit names such as cms_site_brief, cms_get_schema, cms_search, cms_create or cms_create_page_translation. The details are on the AI & MCP page.

MCP, REST API or AI plugin: what's the difference?

All three automate, but not at the same level.

• A REST API is for developers: someone has to write the code that calls it, and every new use case means a new script.
• An AI plugin adds a “generate” button to the editor: handy for a paragraph, limited to whatever the plugin anticipated.
• An MCP server speaks to the agent itself: it discovers the tools, decides which ones to chain, and you talk to it in plain language. “Make the English version of these five pages” becomes a request, not a development project.

The guardrails to insist on

Handing a CMS to an agent only makes sense if you stay in control. Before connecting an assistant, check these five points:

1. User permissions: the agent must never have more power than the person who connected it. Ideally, give it a dedicated account with restricted rights.
2. Drafts by default: what the agent writes must not go live until a human publishes it.
3. Revisions on everything: every change should be traceable and reversible, pages and layouts alike.
4. No immediate deletion: a trash bin rather than permanent removal.
5. Sensitive settings out of reach: domains, sites and users are managed in the interface, not by the agent.

What people actually ask it

• “Here is my client's brief, propose a site structure and write the pages.”
• “The client sent their corrections, apply them page by page.”
• “Add an FAQ and structured data to these pages.”
• “Update the legal notice on all our sites.”

Each time, the work arrives as a draft; the editor reviews, compares and publishes. The site you are reading was built this way, this article included.

How to choose an MCP-ready CMS

An MCP server bolted onto a CMS designed for a visual editor soon hits its limits. Ask these questions:

• Is the content model explicit (schemas the agent can read) or buried in page-builder HTML?
• Is the MCP server built in and maintained with the CMS, or supplied by a third-party plugin?
• Does it cover the whole cycle: reading, writing, translation, media, search?
• Does it respect the CMS's roles and publishing workflow?

Plumixia was designed around these requirements. Book a demo: we connect an assistant in front of you, on your own content.

Frequently asked questions