AI glossary
MCP server
A program that exposes tools, data or prompts to AI applications through the Model Context Protocol.
An MCP server can run on your own computer (a local server, started by the AI app and talking over standard input/output) or on the internet (a remote server, reached over HTTP, often with OAuth sign-in). Examples include servers for GitHub, Google Drive, Slack, databases and web browsers.
Only install servers you trust: a server can read data and take actions with your permissions, and servers that fetch outside content can expose the model to prompt injection.
Example: You add GitHub’s MCP server to Claude Code. From then on you can ask “show me the open issues labelled bug” and the assistant uses that server’s tools to look them up.
In practice
- Prefer the service’s official server or a well-maintained open-source one.
- Give it the least access it needs: read-only tokens when they are enough.
- Remove servers you no longer use.
Step by step in how to install an MCP server.
