Model Context Protocol (MCP)

The Model Context Protocol, or MCP, is an open protocol for connecting AI applications to external systems. An MCP server can expose capabilities such as tools, resources, and reusable prompts to an MCP client.

Tiny example

AI application (MCP client)
        ↓ discovers tools
customer-data MCP server
        ↓ calls approved API
customer system

MCP standardizes the connection and message exchange. It does not automatically make a server, tool, or returned value safe or trustworthy.

For tools, the server advertises a name, description, and input schema. A model may use those definitions to propose a call, but MCP does not require one orchestration pattern. The host controls which connections and tools are available and enforces consent and application policy. The server must independently validate input, enforce access control, rate-limit calls, sanitize output, and execute or reject the tool.

FDE note

Treat every server as an integration boundary. Review authentication, user consent, tool permissions, data exposure, timeouts, and audit logging before enabling it.

Continue with How an MCP-connected assistant chooses tools.