MCP tools: discovery, schemas, and invocation

MCP tools are callable functions a server exposes to a client. A client lists them with tools/list to get each tool's name, description, and input schema, then invokes one with tools/call, passing arguments that fit the schema. The server returns content in the response.

A client sends tools/list and receives an array of tools. Each tool entry describes its name, a description, and an input schema — commonly JSON Schema. This lets a client present the right arguments without prior knowledge of the tool.

The input schema defines the shape of arguments a tool accepts: which fields are required, their types, and their descriptions. The client validates or at least shapes its request against the schema before sending tools/call.

The client sends tools/call with the tool name and arguments. The server executes the operation and returns the result as content. Whether a tool mutates state is determined by the server implementation.

Tools are the primary action surface of MCP. A tool that can write to a database or a file system carries real risk if invoked by an AI agent. Treat tool permissions as a trust boundary: verify what a tool does, what access it has, and whether it should be read-only.