MCP (Model Context Protocol): definition

MCP (Model Context Protocol)

The Model Context Protocol (MCP) is an open standard for connecting AI applications such as Claude or ChatGPT to outside tools and data. A service publishes an MCP server once, and any compatible AI app can use it.

Also called

MCP, Model Context Protocol, MCP server, MCP client

Category

Updated

October 2026

Before MCP, every AI app needed its own integration for every tool. MCP replaces that with one protocol. The documentation compares it to a USB-C port for AI applications. Anthropic introduced it on 25 November 2024, and it is now supported by Claude, ChatGPT, Cursor, Visual Studio Code and many other tools.

It has three parts. The host is the AI application, such as Claude Desktop or Claude Code. The host runs an MCP client for each MCP server it connects to, and the server is the program that exposes a system to the AI. Servers offer three main kinds of building block: tools the AI can call to take actions, resources that supply context such as files or records, and prompts, reusable templates for common tasks. Servers can run locally on your machine or remotely over HTTP. In December 2025 Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation, and said its governance model would stay unchanged.

Why it matters for a SaaS website

MCP decides how much an AI agent can do with the tools your website runs on. If your CMS, site builder or analytics account has an MCP server, an agent can draft pages, update content or pull reports through it, with the permissions you set. If it has none, the agent is limited to what you paste into the chat. When choosing a stack, check which platforms offer servers and what each server can actually change, because coverage varies a lot: some allow edits, others are read-only.

How DesignMe handles it

MCP is part of how we set up AI workflows. We connect chat, the website or repo, the CMS and analytics with separate permissions to read, draft, publish and approve, and every action is logged. In our platform comparisons we note what each server allows. Webflow’s official MCP server lets agents create and style elements, manage CMS items and update SEO settings, and it respects workspace roles and page branching. HubSpot’s remote MCP server gives agents CRM access, but pages and blog posts are read-only there. On code-based Astro or Next.js sites the agent works on the files and opens a pull request, which we cover in Astro vs Webflow.

Example

Say a marketing team runs its site on Webflow. Through the Webflow MCP server, an agent in Claude drafts five new integration pages as CMS items using the existing template, fills in the SEO fields and leaves them unpublished. A content lead reviews the drafts in Webflow and publishes the ones that are ready.

Common mistakes

Assuming MCP means full access. Each server exposes different actions. Read what it can and cannot change.
Connecting everything. Every server is another set of credentials. Connect only what a workflow needs.
Skipping review. An agent with write access should still work in drafts or branches.
Building a custom server first. Check for an official one before writing your own.

Questions

Who owns MCP?

Anthropic introduced MCP in November 2024 and released it as an open standard. In December 2025 it donated the protocol to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. Anthropic said the project's governance model and maintainers would stay the same.

Does my website need an MCP server?

Usually not your own. What matters is whether the tools your site runs on offer one. Webflow has an official MCP server, and code-based sites in Astro or Next.js can be edited by coding agents directly through the repository. Build a custom server only when an agent needs data no existing server provides.