Blog

What MCP (Model Context Protocol) is and what it is for in a school

MCP is the open standard that connects AI assistants such as Claude or ChatGPT to other programs. How it works, what risks it carries and how to evaluate a connector in a school.

Ena by Edena | 8 min read
What MCP (Model Context Protocol) is and what it is for in a school

In the first years of AI assistants, using them with your work data meant copying and pasting: export a list, paste it into ChatGPT, ask for a summary and copy the result back. It was slow, and it also meant taking student and family data out of the system where it is protected. MCP was created to solve that: it is the standard way for an AI assistant to connect directly to another program, with permission, and work with its data without anyone copying anything.

This article explains what MCP is, how it works, what you can do with it in a school, nursery or academy, what risks it carries and how to evaluate a connector before turning it on at your school.

What MCP is, in one sentence

MCP (Model Context Protocol) is an open standard that defines how an AI assistant discovers what it can do in another program, asks for permission to do it and requests data or actions in an orderly way.

That is why it is often called "the USB-C of AI": just as one connector can charge your phone or plug in a screen, one protocol lets any compatible assistant talk to any program that offers an MCP server.

Where it comes from and who uses it

  • November 2024. Anthropic, the company behind Claude, introduces MCP as an open standard.
  • 2025. OpenAI, Google and Microsoft build it into their assistants and developer tools, and it becomes the usual way to connect assistants with other programs.
  • 9 December 2025. Anthropic donates MCP to the Agentic AI Foundation, a fund within the Linux Foundation set up together with Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. According to Anthropic, there were more than 10,000 MCP servers in operation at the time.

The fact that the protocol does not belong to a single company matters to a school: an MCP connector does not tie you to a particular assistant. The same connector works for Claude, for ChatGPT or for whichever tool each person uses.

How it works

Three pieces take part in an MCP connection:

  • The host: the AI application the person uses, such as Claude, ChatGPT or a code editor.
  • The client: the part of the host that maintains the connection with a specific service.
  • The server: what the program you connect to offers, for example the school's management software. It exposes a list of tools, with their descriptions and parameters.

The server can offer three kinds of things:

  • Tools: actions or queries the assistant can request, such as "search students" or "prepare an event".
  • Resources: data or documents the assistant can read as context.
  • Prompt templates: predefined ways of asking for something common.

And there are two ways to connect:

  • Local: the server runs on the person's computer. Common in developer tools.
  • Remote: the server is on the internet and the assistant connects to an address. To get access, the person approves the connection in the service with their own account, through a standard authorisation flow (OAuth), without giving their password to the assistant.

For a school, what matters are remote connectors: nothing needs installing, each person connects their assistant with their own account and the school can withdraw access whenever it wants.

What MCP can do in a school

With an MCP connector to the management software, each person can ask their assistant things like these, always within what their permissions let them see and do:

  • Leadership: "How many new enrolment applications have we had this month and which channel did they come from?"
  • School office: "List the Year 7 families with October bills outstanding."
  • Coordination: "Which teachers have a class on Thursday from 10:00 to 11:00?"
  • Tutors: "Prepare an event for the Year 9B parents' meeting on Thursday 26th at 17:00."
  • IT team: "Which API operation registers a student and which fields does it need?"

The assistant finds the right tool, uses it with the person's permissions, summarises the answer and, if the request involves a change, prepares it and asks for confirmation before carrying it out.

MCP, API and built-in agent: how they differ

  • The API is the software's technical door, so other programs can read and write data. Developers use it to build integrations. We explain it in the article on APIs and webhooks.
  • MCP is a layer designed for AI assistants: it describes the tools in language the model understands and handles the person's permission. Underneath, it usually uses the API itself.
  • A built-in agent lives inside the management software, with its own screens and tools. It needs no connection, but only works inside that software.

They are not mutually exclusive. The same school can have a built-in agent for day-to-day work, an MCP connector for those who prefer working from their assistant and an API for the IT team's integrations. We compare them in the article on AI agents in education.

The risks you should know about

MCP is a protocol; security depends on how each connector is built and how it is used. These are the main risks in a school:

  • Excessive permissions. A connector that runs on an administrator key gives any assistant access to everything. The connector should act as the person who connects it, with their permissions.
  • Prompt injection. Text inside the data, such as a family's message or a file name, can contain phrases the model interprets as instructions. That is why changes should require confirmation and the assistant should treat data as data, not as instructions.
  • Unknown third-party connectors. An unofficial MCP server may ask for more access than needed or do things other than what it describes. Use only the official connector from your software provider.
  • Data going to the assistant's provider. What the assistant reads is processed by the company that offers it. The school must decide which accounts and plans can be used, on what terms and who can connect.
  • Irreversible actions. Issuing invoices, moving money or changing permissions should not be possible from an assistant, even if the person can do it in the software.
  • No record. If there is no trace of what each assistant looked up and did, it cannot be reviewed later.

Ten questions to evaluate an MCP connector

Before turning on a connector at your school, ask for answers to these questions:

  1. Is it the official connector from the software provider?
  2. Does it act as the person who connects it, with their account, or with a shared key?
  3. Does it respect the permissions of each role, just like the software?
  4. Can the school decide which roles may connect it and disconnect everyone at once?
  5. Is it read-only by default? Do changes only appear if the person allows them when connecting?
  6. Does it ask for confirmation before each change and carry it out only once?
  7. Which operations are always excluded: invoices, money, permissions, settings?
  8. Does it hide sensitive data, such as health, bank or identity data?
  9. Does it have usage limits per person and per school?
  10. What does the privacy policy say about what the assistant's provider processes?

Data protection and the AI Act

Connecting an assistant to the school's software is processing personal data, often of minors. Before turning it on:

  • Decide who can use it, by role, and with which assistant accounts: free personal accounts and business or education plans do not have the same terms for the use of data.
  • Review the assistant provider's terms and whether there is a processing agreement covering that use.
  • Tell staff what they can and cannot ask, and that changes are confirmed.
  • Involve your data protection officer. The Spanish data protection authority's guidance on agentic AI, from February 2026, calls precisely for specific governance and the ability to withdraw access immediately.
  • Remember the AI literacy obligation of the EU AI Act, in force since February 2025, which we explain in the guide to the AI Act for schools.

How to do it step by step with ChatGPT or Claude is covered in the guide to connecting your assistant to school data.

Edena's MCP connector

Edena has a remote MCP connector that each staff member can add to their assistant as a custom connector:

  • It works in Claude, ChatGPT, Claude Code, Codex, Gemini CLI, GitHub Copilot, Cursor and any remote MCP client. In Claude it is added from Settings, Connectors.
  • It acts as the person who connects it: when approving access in Edena, they choose the school and whether the assistant only looks things up or can also prepare changes. It sees and does the same as that person in Edena, never more.
  • Six tools: whoami (who it is acting as), search_operations and describe_operation (find and understand the API's operations), read_operation (look things up as the person), and prepare_change and apply_change, which only appear if the person allowed changes: the first shows what will be done and the second does it once, after confirmation.
  • What it never does: issue or cancel invoices, payments, refunds, wallets and direct debit batches; change permissions, the team or the school's settings. Bank, identity and health data and credentials reach it hidden.
  • The school decides: an administrator chooses which roles can connect an assistant and can disconnect them all at once.
  • Limits: up to 120 calls per minute per person and 1,200 per school.
  • Privacy: what the assistant reads is processed by its provider, with the person's account.

Conclusion

MCP is the missing piece for AI assistants to work with a school's real data without copying and pasting lists. Because it is an open standard, the same connector works for whichever assistant each person uses. But its security depends on the connector: acting as the person, respecting their permissions, asking for confirmation, excluding money and permissions, hiding sensitive data and letting the school turn it off are the minimum conditions.

If you want to try it with Edena, the details are in the section on the connector for Claude and ChatGPT. And if you prefer an AI that already lives inside the platform, meet Ena.

Frequently asked questions

What is MCP?
MCP, or Model Context Protocol, is an open standard that defines how an artificial intelligence assistant, such as Claude or ChatGPT, connects to other programs to look up their data or take actions in them. Anthropic introduced it in November 2024 and since December 2025 it has been run by the Agentic AI Foundation, within the Linux Foundation.
What is MCP for in a school?
It lets each staff member ask their AI assistant about the school's real data (attendance, outstanding bills, enrolment applications, the calendar) and, if allowed, prepare changes in the management software, always with their own permissions and confirming before each change.
What is the difference between MCP and an API?
The API is a program's technical door, designed to be used by other programs written by developers. MCP is a layer on top, designed to be used by an AI assistant: it describes which tools exist, what each one does and how to ask for permission, so the assistant can use them from a question in plain language.
Is it safe to connect an AI assistant to school software with MCP?
It depends on the connector. A well-designed one acts as the person who connects it rather than as an administrator, respects their permissions, is read-only unless told otherwise, asks for confirmation before each change, excludes money and permission operations, hides sensitive data and lets the school disable it. Also, what the assistant reads is processed by its provider, so the school must decide which accounts and plans it is used with.
Which assistants support MCP?
The main AI assistants and tools support it: Claude, ChatGPT, Claude Code, Codex, Gemini CLI, GitHub Copilot and Cursor, among others. Remote connectors are usually added from the assistant's settings as a custom connector, with an address and an authorisation in the service being connected.

This content was generated by Ena, Edena's artificial intelligence agent. It may contain errors or inaccuracies and does not constitute legal, tax or professional advice. Edena does not warrant the accuracy, completeness or currency of the information. Consult official sources and, where appropriate, a qualified professional before taking any decision. The cover image is from Unsplash .

The office closes the month without chasing families

Invoices go out with Verifactu, attendance is recorded and each family sees it in the app. Seeing it with a school like yours takes 15 minutes.

Let's talk