Tool Use (Function Calling)

Tool use lets Claude interact with the outside world: database queries, a weather API, sending email, running code. Claude does not run the function itself (for client tools) — it tells you which tool to call and with what input; your code runs it and sends the result back.

7.1 Agentic loop

7.2 Tool definition and manual loop

tools = [{
    "name": "get_order_status",
    "description": "Returns the current status of an order by order ID. Use it when the user tracks an order.",
    "input_schema": {
        "type": "object",
        "properties": {"order_id": {"type": "string", "description": "For example ORD-1234"}},
        "required": ["order_id"],
        "additionalProperties": False,
    },
    "strict": True,            # will match the input schema exactly
}]

messages = [{"role": "user", "content": "Where is my order ORD-1234?"}]
while True:
    r = client.messages.create(model="claude-opus-5-5", max_tokens=4000,
                               tools=tools, messages=messages)
    messages.append({"role": "assistant", "content": r.content})
    if r.stop_reason != "tool_use":
        break
    results = []
    for b in r.content:
        if b.type == "tool_use":
            out = get_order_status(**b.input)          # your function
            results.append({"type": "tool_result", "tool_use_id": b.id, "content": out})
    messages.append({"role": "user", "content": results})   # all results in ONE message

7.3 Tool Runner (SDK helper)

Instead of writing the loop yourself, you can use the SDK's Tool Runner — it runs the request → run tool → send back cycle for you.

from anthropic import beta_tool

@beta_tool
def get_weather(location: str) -> str:
    """Tells the weather of a city.
    Args:
        location: Name of the city, for example Delhi.
    """
    return f"{location}: 31°C, sunny"

runner = client.beta.messages.tool_runner(
    model="claude-opus-5-5", max_tokens=16000, tools=[get_weather],
    messages=[{"role": "user", "content": "Weather in Delhi?"}],
)
for message in runner:      # each iteration is one message; stops when Claude is done
    print(message)

7.4 Important concepts

7.5 Anthropic's built-in tools

r = client.messages.create(
    model="claude-opus-5-5", max_tokens=16000,
    tools=[{"type": "web_search_20260209", "name": "web_search", "max_uses": 5}],
    messages=[{"role": "user", "content": "Tell me this week's AI news"}],
)   # server tool: no loop to write, the result comes in this same response