The Model Context Protocol (MCP) is an open standard for connecting AI applications to outside systems. It gives assistants like Claude and ChatGPT a common way to read data from tools such as calendars, databases and content platforms, and to take actions in them. The official MCP documentation says to "think of MCP like a USB-C port for AI applications."
For businesses, MCP matters because it's becoming one of the main ways AI assistants and agents connect to the rest of the world. Today that mostly means internal workflows, like pulling analytics into a chat. Increasingly, it also means customers' AI tools working with a company's catalog, bookings or software directly.
What is the Model Context Protocol?
Anthropic introduced MCP on November 25, 2024, describing it as "an open standard that enables developers to build secure, two-way connections" between data sources and AI-powered tools. Before MCP, every AI app needed a custom integration for every tool. MCP replaces that with one shared protocol: build a connection once, and any compatible AI app can use it.
Adoption moved quickly. In December 2025, Anthropic donated MCP to the Agentic AI Foundation, a new foundation co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. At the time, Anthropic reported more than 10,000 active public MCP servers and over 97 million monthly SDK downloads, and noted that ChatGPT, Gemini, Microsoft Copilot, Cursor and Visual Studio Code all support it.
OpenAI's Apps SDK, announced in October 2025 for building apps that run inside ChatGPT, also "builds on the Model Context Protocol."
Why MCP matters
It's the plumbing behind AI agents. AI agents need to do more than chat. They need to check inventory, book appointments, look up account data or file a ticket. MCP is one of the most common ways those connections are built.
It creates a new surface for brands. Some companies now offer MCP servers so customers' AI tools can work with their products directly. Shopify, for example, documents MCP-based endpoints that let AI agents search a store's catalog and work with carts. This is one of the building blocks of agentic commerce.
It changes how marketing teams work. A growing number of SEO, analytics, CRM and CMS platforms offer MCP servers. That lets a team ask an assistant to pull a report or draft a page update without jumping between dashboards.
How MCP works
Hosts, clients and servers
MCP uses a client-server design with three participants. The host is the AI application, such as Claude Desktop or Visual Studio Code. The host creates an MCP client for each connection it opens. The server is the program that provides context, such as access to a filesystem, a database or a third-party service.
Servers can run locally on a user's computer or remotely on the web. Local servers typically talk to the host over standard input and output. Remote servers use a transport called Streamable HTTP and usually require authentication, with the documentation recommending OAuth.
Tools, resources and prompts
MCP servers can expose three core building blocks:
- Tools: actions the AI can take, like running a database query or creating a calendar event.
- Resources: data the AI can read, like file contents or records.
- Prompts: reusable templates that help structure a task.
The AI app asks each server what it offers, and the model decides when to call a tool based on the conversation. Each tool comes with a name and a plain-language description, which the model uses to judge whether it fits the user's request.
A living standard
The specification is versioned and still evolving. The current documentation is built around a protocol version dated July 28, 2026, which deprecates some older features. If you build on MCP, check which versions the AI apps you care about support.
MCP in practice: an example
Imagine a hypothetical regional chain of outdoor gear stores. Its marketing team connects an AI assistant to its analytics platform and CMS through MCP. Instead of exporting spreadsheets, the team asks, "Which product pages lost the most organic clicks last month?" and gets an answer drawn from live data.
Later, the company's developers build a small public MCP server that exposes store hours, stock levels and product details. When a shopper's assistant asks whether a particular tent is in stock nearby, it can check the real answer instead of guessing from an old web page. The same facts still live on the website, so search engines and AI crawlers see a consistent picture.
How to prepare for MCP as a marketer
- Get your public facts right first. MCP doesn't replace your website. AI tools still learn about you mostly through crawling and search, so accurate, crawlable pages and schema markup come first.
- Audit the tools you already use. Check whether your analytics, CRM, SEO and CMS platforms offer official MCP servers your team could use.
- Decide whether customers' agents need access. If buyers would benefit from an assistant checking pricing, availability or bookings, talk to your developers about whether an MCP server makes sense.
- Write clear tool descriptions. If you do build one, tool names and descriptions work a bit like metadata. Models use them to decide when to call your tool.
- Involve security early. MCP's own security best practices cover risks like over-broad permissions and compromised local servers. Only connect servers from sources you trust.
Common mistakes and misconceptions
"MCP is an Anthropic-only technology." Anthropic created it, but it's an open standard now housed in an independent foundation and supported across OpenAI, Google, Microsoft and many other companies' products.
"An MCP server will get my brand into AI answers." Not on its own. MCP connects tools that a user or app has chosen to enable. It isn't a ranking or citation system, and most AI search answers still draw on crawled and indexed web content.
"MCP replaces APIs." MCP usually sits on top of existing APIs. It standardizes how AI apps discover and call them.
"Any MCP server is safe to install." Local servers can run code on your machine, and remote ones may get access to your accounts. Treat them like any software you install, and grant only the permissions they need.
MCP vs. llms.txt
Both come up in conversations about making sites AI-friendly, but they do different jobs. llms.txt is a proposed plain-text file that summarizes a site's content for language models to read. MCP is a protocol that lets AI apps fetch live data and take actions through a server. One is a static guide; the other is a working connection.
How Butter approaches MCP
MCP isn't a separate Butter service, and like broader agent readiness, it overlaps heavily with work we already do. Our GEO plan covers AI prompt testing, indexing strategy and schema markup guidance, which help AI systems find and read your key facts, however they reach you. When we review a site, we flag pricing, service details or forms that an agent would struggle with, and include those in our on-page recommendations.
If your team decides an MCP server is worth building, that's a developer project, and the same rule applies: the facts it exposes should match your website. You can see how our work fits together on our GEO service page.
Frequently asked questions
Who created the Model Context Protocol?
Anthropic introduced MCP in November 2024. In December 2025, Anthropic donated it to the Agentic AI Foundation, which it co-founded with Block and OpenAI, so the standard is now developed in the open with broad industry support.
Does MCP help with SEO or GEO?
Not directly. MCP connects AI apps to tools and data that a user has enabled. It doesn't influence rankings or citations in AI search, which still rely mainly on crawled and indexed web content. It can matter for brands whose customers use AI agents to shop or book.
Do I need an MCP server for my business?
Most small and mid-sized businesses don't yet. It's worth considering if customers would benefit from AI assistants checking live information, like stock, pricing or appointments, or if your product is software people want to use from inside AI tools.
What is the difference between an MCP client and an MCP server?
The MCP server is the program that provides data or actions, such as a connection to a database or a software tool. The MCP client lives inside the AI application and maintains the connection to that server on the app's behalf.
Sources and further reading
- Anthropic: Introducing the Model Context Protocol
- Anthropic: Donating the Model Context Protocol and establishing the Agentic AI Foundation
- Model Context Protocol: What is MCP?
- Model Context Protocol: Security best practices
- OpenAI: Introducing apps in ChatGPT and the new Apps SDK
- Shopify: Storefront MCP