What is an MCP server?

MCP stands for Model Context Protocol, an open standard for connecting AI assistants to other software. An MCP server is a connector built on that standard. Once it's set up, your assistant can look things up in, or take actions on, the system it connects to, instead of relying only on what you paste into the chat.

A plain-language example

Without a connector, you export a task list from your project tool, paste it into the chat, and ask for a status summary. With a connector to that tool, you ask for the summary and the assistant fetches the current tasks itself.

Common kinds of connections include:

• Files and folders on your computer
• Project and task trackers
• Documents, spreadsheets, and calendars
• Databases or internal systems, through a connector someone builds

Which connectors exist, and which assistants support them, changes quickly. Check the documentation for your AI tool and the system you want to connect.

Where it fits in the book

MCP servers appear in Chapter 12, Automation and Integration, alongside Python scripts. That's one of the technical chapters. Chapters 1–8 need no coding, and you can get a lot of value without ever setting up MCP.

Before you connect anything

• Get approval. A connector gives the AI access to real systems and data, so check your organisation's policy and involve IT where needed.
• Start read-only. Let the assistant look before you let it change anything.
• Use trusted sources. Only install connectors from sources you or your IT team trust.
• Review actions. Treat anything the assistant proposes to change as a draft to approve.

For a friendly explanation, read Don't Understand MCP Servers and Integrations and Can't Integrate AI With Existing Project Tools.