All entries by category
Concept · MCP & Tooling
A function the harness exposes for the agent to call, and the only way the agent perceives and acts on its environment.
The model's output naming a tool and its arguments: structured text that the harness has to read and execute.