What Is MCP (Model Context Protocol)? A Plain-English Guide for Business

MCP (Model Context Protocol) is an open standard that lets AI apps like Claude or ChatGPT connect to outside systems. Those systems include your files, databases, CRM, accounting software and other business tools. The official MCP documentation compares it to a USB-C port for AI apps: one standard way to plug things in.

Before MCP, every AI app needed its own custom connector for every tool. MCP replaces that tangle with one shared set of rules. A tool vendor builds one MCP server, and any AI app that speaks MCP can use it.

This guide explains what MCP is, where it came from, who controls it now, and what it means for a small or mid-size business. No coding knowledge is needed.

What MCP means in one sentence

The MCP intro page defines it as "an open-source standard for connecting AI applications to external systems." In plain terms, it is a common language. AI apps use it to ask other software what it can do, and then to ask it to do those things.

Think about what an AI assistant can do on its own. It can write and reason, but it cannot see your sales pipeline, last month's invoices or your website analytics. MCP is the bridge that lets the AI read that context and take approved actions.

Where MCP came from

Anthropic announced MCP on November 25, 2024. The company said even capable models were limited by their isolation from data. Every new data source needed its own custom integration, which made connected systems hard to scale.

The first release included three things:

• The MCP specification and software development kits (SDKs)
• Local MCP server support in the Claude Desktop apps
• An open-source repository of example MCP servers

Anthropic named Block and Apollo as early adopters. It also shipped ready-made servers for Google Drive, Slack, GitHub, Git, Postgres and Puppeteer.

Who owns MCP now

No single AI company controls MCP today. On December 9, 2025, Anthropic donated MCP to the Agentic AI Foundation (AAIF), a directed fund under the Linux Foundation. Anthropic, Block and OpenAI co-founded the foundation. Google, Microsoft, AWS, Cloudflare and Bloomberg backed it.

At the time of the donation, the MCP project blog reported more than 10,000 active public MCP servers. It also reported more than 97 million monthly SDK downloads across Python and TypeScript. For the full governance story, see who owns MCP.

How MCP works, in brief

MCP uses a simple client-server design. The specification names three roles:

Servers offer three main building blocks:

• Tools: actions the AI can take, such as "create invoice" or "search tickets."
• Resources: read-only context, such as a file, a database schema or a report.
• Prompts: reusable templates a user can pick, such as "summarize my week."

Messages travel as JSON-RPC 2.0, a lightweight format for calling functions between programs. A server can run on your own computer or as a hosted service on the internet. Our guide on how MCP works walks through a full request step by step.

What changed in the July 2026 specification

MCP keeps evolving. The 2026-07-28 changelog made the protocol stateless. Every request now carries its own version and capability details, so servers no longer need a start-up handshake or a stored session.

The same release moved some features into optional extensions. These include Tasks for long-running jobs, Skills over MCP for packaged workflow instructions, and MCP Apps for interactive charts and forms inside a chat.

What MCP can do for a business

The official docs list examples such as an agent that reads your Google Calendar and Notion, or an enterprise chatbot that queries several databases. In business terms, MCP lets an AI assistant:

• Pull numbers from analytics, accounting or CRM tools without copy and paste
• Draft records in your systems, such as a task, invoice or support note
• Combine context from several tools in one conversation
• Work inside your existing permissions when the server is built that way

Many software vendors now publish their own MCP servers. Examples include GitHub, Stripe and HubSpot. See 25 real MCP examples by department for more.

Which AI apps support MCP

According to the MCP intro page, supporting apps include Claude, ChatGPT, Visual Studio Code and Cursor, among many others. Support levels differ by app and by plan. Some apps support remote servers, some only local ones, and extensions vary. Check our list of AI apps that support MCP before you commit.

What MCP is not

MCP is a protocol, not a product you buy. It does not replace your software's API. Most MCP servers sit on top of an existing API and translate it for AI apps. Our comparison of MCP vs API explains the difference.

MCP also does not make an AI trustworthy by itself. The specification says plainly that MCP "cannot enforce these security principles at the protocol level." The apps and servers that implement it are responsible for consent and access control.

Risks to understand before you connect

The specification treats tools as "arbitrary code execution" and says hosts must get explicit user consent before invoking any tool. That tells you how seriously to take a connection.

Practical steps for a business:

• Connect only servers from vendors you trust, ideally their official server.
• Grant the narrowest access you can, such as read-only where it is offered.
• Keep tool-approval prompts turned on, especially for actions that move money or delete data.
• Remember that text inside your data can try to trick an AI (prompt injection). Stripe's MCP docs warn about this directly.

Read is MCP safe? for a plain-English risk overview.

Next step

If you want to try MCP, start with one or two official servers for tools you already use. Our checklist for choosing MCP servers shows what to check before you connect anything. You can also browse the full MCP Guide.

Frequently asked questions