What Is an MCP Server?
An MCP server acts as the bridge between an AI application and the systems it needs to interact with, exposing capabilities to any MCP-compatible client, such as an AI coding assistant or a desktop AI assistant. It receives structured requests from the model, like a file read, a function call, or a shell command, executes them within a defined environment, and returns the results in a standardized format. A single server can expose tools (functions the model can execute), resources (data the model or user can access), and prompts (templated workflows), which means its security posture matters as much as the model calling it: an overpermissioned or poorly validated server gives a model more reach than intended.
Why MCP Servers Matter
- This is the execution layer, where an AI model's requests turn into actual actions on real systems.
- Each server can expose tools, resources, and prompts, all of which need their own access controls, not a single blanket permission.
- A vulnerability in an MCP server is as exploitable as a vulnerability in the model calling it, and often easier to miss during review.
