Get Started with Anthropic Tool Use for Claude

1. 1
   Set Up Your Environment

   Install the official Python library and configure your API key. The client will automatically read the key from the `ANTHROPIC_API_KEY` environment variable.
2. 2
   Define Your Tools

   Create a list of tools Claude can use. Each tool needs a name, description, and an `input_schema` following the JSON Schema format. This structure tells Claude what the tool does and what arguments it expects.
3. 3
   Make the Initial API Call

   Send your prompt and the tool definitions to Claude. If the model determines a tool is needed, the response will stop and include a `tool_use` content block instead of a final answer. This block contains the tool name and arguments.
4. 4
   Execute the Tool and Get the Final Answer

   Parse the `tool_use` block from the response. Execute your corresponding local function with the provided arguments. Then, using Anthropic Tool Use, send the tool's output back to Claude in a new message to get a final, natural language response.

import os
import json
import anthropic

# Assumes ANTHROPIC_API_KEY is set as an environment variable
client = anthropic.Anthropic()

def get_weather(location):
    """Mock function to get weather. In a real app, this would call a weather API."""
    if "san francisco" in location.lower():
        return json.dumps({"location": "San Francisco", "temperature": "72", "forecast": "Sunny"})
    elif "tokyo" in location.lower():
        return json.dumps({"location": "Tokyo", "temperature": "65", "forecast": "Cloudy"})
    else:
        return json.dumps({"location": location, "temperature": "unknown"})

# Define the available tools
tools = [
    {
        "name": "get_weather",
        "description": "Get the current weather for a specific location.",
        "input_schema": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "The city and state, e.g., San Francisco, CA"
                }
            },
            "required": ["location"]
        }
    }
]

user_prompt = "What's the weather like in San Francisco?"
print(f"\n👤 User: {user_prompt}")

# 1. First API call: Send the prompt and tools to Claude
message = client.messages.create(
    model="claude-3-sonnet-20240229",
    max_tokens=4096,
    messages=[{"role": "user", "content": user_prompt}],
    tools=tools,
    tool_choice={"type": "auto"}
)

print("\n🤖 Assistant: Initial response")
print(message)

# 2. Check if the model wants to use a tool
tool_calls = [block for block in message.content if block.type == "tool_use"]

if not tool_calls:
    print("\n🤖 Assistant: No tool use requested. Final answer:")
    print(message.content[0].text)
else:
    print("\n🛠️ Tool call(s) requested:")
    tool_results = []
    for tool_call in tool_calls:
        tool_name = tool_call.name
        tool_input = tool_call.input
        tool_use_id = tool_call.id
        print(f"  - Tool: {tool_name}, Input: {tool_input}")

        # 3. Execute the tool
        if tool_name == "get_weather":
            tool_output = get_weather(location=tool_input.get("location"))
            tool_results.append({
                "type": "tool_result",
                "tool_use_id": tool_use_id,
                "content": tool_output
            })

    print("\n🔄 Sending tool results back to the model...")

    # 4. Second API call: Send tool results back to the model
    response_message = client.messages.create(
        model="claude-3-sonnet-20240229",
        max_tokens=4096,
        messages=[
            {"role": "user", "content": user_prompt},
            {"role": "assistant", "content": message.content},
            {"role": "user", "content": tool_results}
        ],
        tools=tools
    )

    print("\n🤖 Assistant: Final response after tool use:")
    print(response_message.content[0].text)