The Tool-Call Round Trip

Tool calling is a two-turn dance. You send the user's message along with your tool schemas; the model answers with a tool_use block saying "call get_weather with {city: 'Paris'}"; your code runs that function and sends the answer back as a tool_result; then the model uses it to write the final reply. This lesson builds the middle step -> the part where the model's request becomes a real Python call and the result is packaged for the trip back.

Here is the real loop (illustrative only -> we do not call the network):

resp = client.messages.create(model="claude-sonnet-4-6", max_tokens=1024,
                              tools=TOOLS, messages=history)
# resp.stop_reason == "tool_use"; find the tool_use block:
for block in resp.content:
    if block.type == "tool_use":
        result = TOOLS_BY_NAME[block.name](**block.input)
        history.append({"role": "assistant", "content": resp.content})
        history.append({"role": "user", "content": [
            {"type": "tool_result", "tool_use_id": block.id, "content": str(result)}
        ]})
final = client.messages.create(model="claude-sonnet-4-6", max_tokens=1024,
                               tools=TOOLS, messages=history)

The two non-negotiable details: you call the function the model named with the arguments it gave, and you echo its tool_use_id back in the tool_result so the model can match the answer to the request. We model the response as a plain dict so it runs offline:

{"content": [{"type": "tool_use", "id": "toolu_1",
              "name": "get_weather", "input": {"city": "Paris"}}],
 "stop_reason": "tool_use"}

You are building run_tool_call(resp, tools), where tools maps a tool name to a Python function. It must:

• Find the first block in resp["content"] whose type is "tool_use".
• Look up tools[name] and call it with the block's input as keyword arguments: tools[name](**block["input"]). If the name is not in tools, raise (a KeyError or ValueError) -> never run a tool the model hallucinated.
• Return the follow-up user message that carries the result back:

{"role": "user",
 "content": [{"type": "tool_result",
              "tool_use_id": <the block's id>,
              "content": str(result)}]}

Stringify the result with str(result) -> the tool_result content travels as text. Build it so two different tool calls (different name and input) produce two different tool_result messages, the original tool_use_id is echoed back, and an unknown tool name raises.