What Is MCP?
It has since been adopted well beyond its original creator and is now supported across much of the AI industry, making it a de facto standard for how agents and applications access context and take action. Before MCP, connecting an LLM to a new tool or data source meant writing a custom integration for that specific combination. MCP standardizes that connection so any MCP-compatible model can talk to any MCP-compatible tool without bespoke glue code.
The Three Components
- Hosts — the AI application using MCP (an AI coding assistant, a desktop AI assistant, an agent framework).
- Clients — the component within the host that manages communication with MCP servers.
- Servers — the external services that expose tools, data, and capabilities to the host through MCP.
Why MCP is a Security Consideration
- Each MCP server a model connects to is a new integration point, its security posture matters as much as the model calling it.
- An overpermissioned or poorly validated MCP server gives a model more reach than intended.
- A single client can connect to multiple servers simultaneously, meaning a successful attack against one tool can potentially chain across every other tool the agent has access to.
