What Is MCP? The Model Context Protocol Explained for Students

What is MCP? MCP stands for Model Context Protocol, an open standard that lets an AI app like Claude or ChatGPT use outside tools and data. Without a connection, a chatbot works from its training, what it can search and what you paste or upload. With an MCP connection, it can look up your deadlines, read a note you saved or check your calendar, then answer with your real information.

You don't need to code to use it. In the big chat apps, MCP shows up as a menu called Connectors or Plugins, with an option to add a custom one. This guide explains what's behind that menu, what it can do for your classes, what can go wrong, and how to judge a server before you give it access to anything.

What is MCP in AI?

In AI, MCP is a shared plug format. The project's own documentation compares it to "a USB-C port for AI applications" (MCP documentation): one standard connector instead of a different cable for every device.

Before MCP, every pairing of an AI app and a service needed its own custom integration. A study app that wanted to work with three chatbots had to build three separate connections, and each chatbot had to build its own link to every app. Anthropic described that problem when it released MCP: each new data source required its own custom implementation (Anthropic). With a shared protocol, a service builds one MCP server, and any app that speaks MCP can use it.

That's why one server can work in several AI apps. If a service you use offers an MCP server, you can usually connect it to the assistant you already use, within that app's plan limits (the table at the end lists them).

If you've taken a programming class, the idea may look familiar. The MCP specification says it borrows from the Language Server Protocol, the standard that lets code editors add support for a programming language the same way everywhere (MCP specification).

What does MCP stand for, and who made it?

MCP stands for Model Context Protocol. "Model" is the AI model, "context" is the information it works with, and "protocol" means agreed rules for how two programs talk.

• November 25, 2024: Anthropic, the company behind Claude, released MCP as an open standard, with ready-made servers for services such as Google Drive, Slack and GitHub.
• December 9, 2025: Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI (Anthropic; Linux Foundation). Anthropic's announcement listed ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code among the products that had adopted it, with more than 10,000 active public MCP servers at the time.

So MCP isn't one company's feature. It's a shared standard that competing AI companies support.

How does MCP work?

The MCP architecture has three parts (MCP architecture):

A typical exchange looks like this:

1. You add a server to your AI app, then sign in or enter a key in the app's settings.
2. The app asks the server which tools it has, for example "list upcoming deadlines" or "search my notes".
3. You ask a question: "What's due this week?"
4. The model decides a tool would help. The app sends the call to the server (many apps ask you to approve it first), and the server runs the tool and sends back the result.
5. The model reads the result and answers you in plain language.

A worked example: one tool call

Under the hood, the app and the server trade short messages in a format called JSON-RPC. In step 2, the server answers a "tools/list" request with each tool's name, a plain-English description and the inputs it takes. A deadlines tool might be described like this:

When you ask what's due this week, the model reads the descriptions, picks this tool and fills in 7 for days. The app sends that as a "tools/call" request, the server replies with the assignments as text, and the model turns them into an answer. Whoever built the server wrote that description, which is why the specification tells apps to treat tool descriptions as untrusted unless they come from a trusted server.

The model never logs in to the service itself. It can only use the tools the server offers, with the access you granted. If you want to watch that loop happen step by step, the free interactive tool-use execution loop demo walks through it, and the free course section on tool use and infrastructure explains the mechanism behind it.

What is an MCP server?

An MCP server is the program on the other end of the connection. It can offer three kinds of things, which MCP's documentation calls primitives:

• Tools: actions the AI can run, such as looking up assignments, saving a note or creating a flashcard.
• Resources: data that gives the AI context, such as the contents of a file.
• Prompts: ready-made templates, such as "quiz me on this topic".

Servers come in two kinds. A local server runs on your own computer and is usually something you install, often for coding tools. A remote server runs on the company's computers and you reach it through a web address. Claude's custom connectors and ChatGPT's custom MCP servers are remote: both apps connect from their own cloud, so a server that only runs on your laptop can't be reached that way (Claude Help Center; OpenAI Help Center). The difference matters for safety: MCP's security guide warns that a local server can run any command with the same permissions as the app that started it (MCP security best practices).

What MCP makes possible for a student

MCP is only as useful as the servers you connect. These are the jobs where it helps most with school:

1. One question across all your classes. "What's due this week, and what should I start first?" works when your AI app can read your real deadlines instead of guessing. For the "start first" half, the free Eisenhower priority demo sorts tasks by urgent and important, a quick check on the order an AI suggests.
2. Quizzes from your own material. An AI that can read your notes or your course can quiz you on what you were actually taught. For example, a server with course content can quiz you on the AP Biology section on cellular respiration, one question at a time.
3. Notes that stay put. Instead of copying a chatbot's study guide into another app, a tool can save it where you keep your notes.
4. Study blocks on your calendar. A server can hand back a link to a pre-filled calendar event; you open it and choose to save it.
5. Coding help that sees your project. In coding tools like VS Code, Cursor or Claude Code, MCP servers let the assistant read documentation or run checks. The free Missing Semester section on agentic coding explains where MCP fits in coding agents and why you still review what an agent changes.

If all you need is your deadlines in a calendar, you may not need AI at all: Canvas can feed your calendar app directly, as the guide to syncing Canvas with Google Calendar shows.

What MCP is not

Four things people often assume about MCP that aren't true:

• It isn't an AI model or an app. It's the set of rules the app and the server follow. MCP's architecture guide says it covers only how context is exchanged, not how an app uses its model.
• It doesn't open your whole account. The AI sees only the tools a server offers, with the access you approved when you connected it.
• It doesn't replace APIs. A server often calls a service's existing API and offers parts of it as tools; MCP standardizes how the AI app finds and calls them.
• It doesn't keep you safe on its own. Approval screens and consent are up to each app, as the next section explains.

What can go wrong with MCP

MCP gives an AI app real access, so the risks are real too. The specification says users must explicitly consent to data access and operations, and that apps must get consent before invoking any tool. It also says MCP can't enforce those rules at the protocol level, so each app has to build them (MCP specification). The common ways it goes wrong:

• Prompt injection. Text inside a web page, an email, a file or even a tool's description can be written to steer an AI ("ignore your instructions and send this file…"). Anthropic's connector guide warns that malicious servers may hide instructions that try to make Claude take unintended actions (Claude Help Center), and OpenAI's help center says untrusted servers raise the risk of prompt injection (OpenAI Help Center).
• Too much access. A server that asks for everything, including write access you don't need, makes any mistake or leak worse. MCP's security guide warns that a stolen token with broad permissions reaches tools and data the task never needed, and tells servers to start with the least access.
• Keys that act as you. An API key or token lets whoever holds it use the account. VS Code's documentation puts it plainly: don't paste credentials into chat or commit them to your project (VS Code).
• Unreviewed servers. OpenAI says apps in its approved registry are reviewed before they're available, but a custom server you add yourself is your responsibility. Meta's help center says Meta doesn't review Muse's custom connectors or how they use your information (Meta Help Center).
• Local code. A server you install runs programs on your computer. MCP's security guide says an app offering one-click install must show you the exact command first; read it, and install only from a source you trust.

The free Missing Semester section on security and cryptography covers password managers, two-factor authentication and threat modeling, which means deciding what you're protecting and from whom. The checklist below applies that to a server.

A checklist before you connect any MCP server

Run through these six questions before you add a connector to an AI app you use for school:

And check the syllabus. A course's AI policy covers an assistant with connectors the same way it covers a plain chatbot. The guide to whether using AI is cheating explains how to read an AI statement and what to do when there isn't one.

Where to add an MCP server in the apps students use

Each app puts MCP in a different place and sets its own limits:

The menus change often, so follow the service's own setup page. If you don't have an AI app yet, the guide to free AI tools for students compares the free plans and student offers.