Tool use (function calling)

What it is

With tool use, you describe the available tools to the model: name, purpose and input fields. The model can reply with a request to call one. OpenAI calls this a multi-step conversation between your application and the model. Anthropic distinguishes client tools, which your application runs, from server tools such as web search, which Anthropic runs for you.

How it works

1. Define tools with a name, description and input schema.
2. The model decides a tool would help and returns a structured call with arguments.
3. Your application runs the tool, sends back the result, and the model continues.

Worked example. A support assistant has a tool lookup_order(order_id). A customer asks, “Where’s order 4471?”

1. The model returns a call: lookup_order with order_id: "4471".
2. Your code checks the customer owns that order, queries your store’s API and returns “shipped, arriving Thursday”.
3. The model writes the reply.

A second tool, refund_order, would need a human approval step before your code runs it.

Benefits

• Connects models to live data and real systems.
• Structured arguments are easier to validate than free text (see structured outputs).
• The foundation of MCP and agent frameworks.

Limits and risks

• The model may pick the wrong tool or the wrong arguments, so validate inputs.
• Actions with side effects need approval steps.
• Tool results can carry prompt injection.