docs.claude.comClaude tool use documentation
Tool calls fail. How your agent reacts to failures determines whether it's reliable or brittle.
Modern LLMs can request multiple tool calls in one step. Here's when to use parallel calls and how they change agent latency.
Without budgets, an agent can run away: thousands of tool calls, exhausted APIs, massive bills. Here's how to stop that.
The description field of your tool IS the prompt to your agent. Here's how to write descriptions that produce reliable tool use.
Well-designed tools are the difference between an agent that works and one that flails. Here's what actually matters.