MCP explained for people who don't write code

MCP, the Model Context Protocol, is an open standard that lets AI tools like Claude and ChatGPT connect to other software, such as your calendar, docs, project tracker or database. Instead of copying information into a chat, you connect a tool once, and the AI can read from it or take actions in it when you ask.

You do not need to write code to use MCP. You do need to understand what you are connecting, what it can do, and who approved it. This guide explains MCP in plain terms.

What MCP is, in one paragraph

The official site describes MCP as "an open-source standard for connecting AI applications to external systems" and compares it to a USB-C port: one standard shape that many devices can plug into. Without a shared standard, each AI product has to build its own integration with each piece of software. With MCP, a company can build one connector for its software, and any AI tool that supports MCP can use it. Anthropic created the standard, and the official site lists support in Claude, ChatGPT, VS Code, Cursor and many other tools.

How it works, without the jargon

There are two sides to every MCP connection.

The AI app. This is the tool you talk to: Claude, ChatGPT, Cursor, VS Code, or a custom agent. The MCP docs call this the host.
The MCP server. This is the connector on the other side. It sits in front of a piece of software, such as a project tracker, and tells the AI app what it offers. Despite the name, it is often just a web address you paste into a settings screen.

When you ask a question, the AI app checks which connectors are available, decides whether one can help, and asks it for information or an action. The connector does the work and sends back a result.

Servers can be local or remote. A local server runs on your own computer, for example to read files in a folder. A remote server runs on the internet, usually hosted by the software company, and you sign in to it like any other app.

The three things a connector can offer

The MCP docs define three building blocks a server can expose. Here they are with everyday examples:

Building block	What it is	Example
Tools	Actions the AI can take	Create a ticket, search a database, send a message
Resources	Information the AI can read	A file's contents, a customer record, a database schema
Prompts	Reusable templates for common tasks	A "summarize this incident" template with the right format

Tools are the part to watch. Reading data is one thing. Creating, editing or deleting it is another, and that is where approvals matter.

Where you will see MCP today

You will rarely see the letters "MCP" in an app. Most products call these connections connectors or apps. Some current examples, from each vendor's own docs:

Claude

Claude supports custom connectors using remote MCP on Free, Pro, Max, Team and Enterprise plans, with Free limited to one custom connector. On Team plans, only owners can add custom connectors for the organization. Members then find the connector in their settings and click Connect to sign in. Claude shows approval requests before it uses a tool, and you can choose "Allow always" for tools you trust.

ChatGPT

OpenAI's help center describes a developer mode for MCP apps in ChatGPT. Full MCP support, including actions that change data, is in beta for Business, Enterprise and Edu workspaces, and a workspace admin has to turn developer mode on. Once an admin approves an app, ChatGPT uses a fixed snapshot of its tools until an admin reviews and publishes an update.

Coding tools and Gemini

Cursor and VS Code both support MCP servers. Google's Gemini CLI does too, and by default asks for confirmation before it runs a tool unless you mark a server as trusted.

A worked example

Say your team tracks product decisions in a project tracker that offers an MCP server. An admin adds it as a connector in Claude. Now a product manager can ask:

"What did we decide about the bulk export feature, and is anyone working on it?"

Claude checks its connectors, sees the tracker, and calls a search tool. It reads the matching tickets and the decision note, and answers with links. If the PM then says "create a ticket to update the docs," Claude asks for approval before calling the create-ticket tool, unless someone already chose "Allow always" for it.

Without MCP, the PM would open the tracker, search, copy the relevant bits into the chat, and paste the answer back out.

What MCP does not do

MCP is plumbing, and plumbing has limits:

It does not decide what is true. If your tracker has three conflicting tickets about bulk export, the AI gets all three. The MCP docs state plainly that the protocol "does not dictate how AI applications use LLMs or manage the provided context."
It does not pick the right context for you. Connecting everything is not the same as giving an agent the five facts it needs. That is context engineering.
It does not make a connector safe. A connector can do whatever its tools allow.

Before you connect anything: a checklist

Anthropic warns that custom connectors can reach services it has not verified, and that a malicious server can attempt prompt injection. Use this list before you add a connector:

Who built it? Prefer connectors from the software vendor itself or from your own team.
What can it do? Read the list of tools. Note any that create, edit, delete or send.
What can it see? Check the permissions you grant when you sign in. Narrow them if you can.
Who approved it? On team plans, an admin should own the list of approved connectors.
When do you approve actions? Keep approval on for anything that changes data. Use "Allow always" only for tools you trust to run unsupervised.
Who reviews changes? Connectors can change their tools over time. Someone should check.

For a fuller treatment, see guardrails for AI agents.

FAQ

Do I need to write code to use MCP?

No. Many products now offer MCP connectors you add by pasting a URL or picking from a directory in your AI app's settings. Building your own MCP server does require code.

Is MCP only for Claude?

No. Anthropic created it, but it is an open standard. ChatGPT, Cursor, VS Code, Gemini CLI and many other tools support it.

What is the difference between MCP and an API?

An API is how one piece of software talks to another, and every API is different. MCP is a common layer on top, so an AI tool can discover and use many services in the same way without a custom integration for each.

Is MCP safe?

It is as safe as the connectors you choose and the permissions you grant. Use trusted connectors, grant narrow access, and keep approvals on for actions that change data.

The short version

MCP is a common plug that lets AI tools reach the rest of your software. It removes a lot of copy-paste, and it makes agents more useful. It also moves the hard questions somewhere new: which connectors you trust, what each one can do, and whether the information behind it is current. If your goal is to give agents your company's context, MCP is one route, covered in how to give AI agents your company context.

We're building Shane to give every AI tool and agent your team's context and decisions, with sources and permissions, so every connection draws on the same current facts. If that's a problem your team has, you can join the waitlist at tryshane.ai.