AI Agents & Tool Use · Lesson 2

Function calling in Claude

The full tool_use loop in Claude: tool_use and tool_result blocks, parallel calls, and forced tool selection via tool_choice.

30 min read4 questions in quizReady prompt includedIn progress

Practical exercise

What to do after this lesson

Implement the full loop: request → tool_use → run function → tool_result → final answer. Add handling for two parallel tool_use blocks in one reply. Return a separate tool_result for each in a single user message.

Task grader

Implement the full loop: request → tool_use → run function → tool_result → final answer. Add handling for two parallel tool_use blocks in one reply. Return a separate tool_result for each in a single user message.

Your answer

Ready-to-use prompt

Template for this lesson

Copy and adapt to your context. Text in angle brackets should be replaced.

Help me implement a tool use loop on Claude (anthropic Python SDK).
Tools: <list with input_schema>
Special case: <parallel calls / forced tool / extraction>

Give working code: a while loop on stop_reason == "tool_use",
accumulating messages, correctly binding tool_result to tool_use_id,
handling is_error. No frameworks.

The tool_use loop in Claude

Claude returns stop_reason: "tool_use", and a tool_use block appears in content with id, name, input. You run the function and continue the same conversation, appending a role: "user" message with a tool_result block.

import anthropic client = anthropic.Anthropic() tools = [ { "name": "get_weather", "description": "Get current weather for a city. Use when the user asks about weather.", "input_schema": { "type": "object", "properties": { "city": {"type": "string", "description": "City name"}, "units": {"type": "string", "enum": ["celsius", "fahrenheit"]} }, "required": ["city"] } } ] messages = [{"role": "user", "content": "What's the weather in Paris?"}] response = client.messages.create( model="claude-sonnet-4-5", max_tokens=1024, tools=tools, messages=messages, ) if response.stop_reason == "tool_use": tool_block = next(b for b in response.content if b.type == "tool_use") result = run_weather(tool_block.input["city"]) # your code messages.append({"role": "assistant", "content": response.content}) messages.append({ "role": "user", "content": [{ "type": "tool_result", "tool_use_id": tool_block.id, "content": result, }], }) final = client.messages.create( model="claude-sonnet-4-5", max_tokens=1024, tools=tools, messages=messages, ) print(final.content[0].text)

Key point: you return the entire assistant response.content (it may contain text + several tool_use blocks), and each tool_result references a tool_use_id.

Forcing a choice: tool_choice

tool_choice={"type": "tool", "name": "get_weather"} # must use this tool tool_choice={"type": "any"} # any, but definitely a tool tool_choice={"type": "auto"} # default: the model decides

tool_choice: "any"/"tool" disables the text answer — useful for extraction tasks where you want strictly structured JSON. Return a function error as a tool_result with "is_error": true — Claude will try to recover.

Report a bug

Function calling in Claude

Task grader

Prompt sandbox

Quiz — 4 questions

Discussion

The tool_use loop in Claude

Parallel calls

Forcing a choice: tool_choice