Module 1 · What agents are · Lesson 2
Tools and the agent-computer interface
How tool calling works on the wire, how to design tools a model uses correctly, and how MCP lets one tool server work in many harnesses.
By the end of this lesson you can
- Describe the request-response cycle of a tool call, including tool definitions, tool_use and tool_result blocks
- Design a tool's name, description, input schema, output and error messages for an agent rather than a human programmer
- Apply mistake-proofing (poka-yoke) to a tool you design, and estimate the token cost of a tool's output
- Explain what MCP standardises, and connect a TypeScript tool to Claude Code as a script or an MCP server