Tool Use and Function Calling: Give Claude Hands to Reach the Real World

TL;DR: Tool use is the mechanism that gives Claude the ability to call your functions — define a JSON schema, send it with your message, and Claude returns a structured call your code executes before resuming the conversation.

What you'll be able to do after this:

• Define a custom tool (a JSON Schema function signature) that Claude can invoke during a conversation
• Handle the two-request round trip: receive a tool_use block, run your function, send back a tool_result
• Force Claude to call a specific tool with tool_choice and add strict: true for schema-guaranteed inputs

How the round trip works

Tool use is a two-turn exchange. Turn 1: you send a message + a tools list. Claude decides to call one and responds with stop_reason: "tool_use" and a tool_use block containing the function name and its arguments. Turn 2: your code executes the function, then you send the result back in a tool_result block. Claude picks up the conversation and answers using the real data.

Python walkthrough

import anthropic

client = anthropic.Anthropic()

tools = [{
    "name": "get_weather",
    "description": "Return current weather for a city.",
    "input_schema": {
        "type": "object",
        "properties": {
            "location": {"type": "string", "description": "City and state, e.g. Austin, TX"}
        },
        "required": ["location"]
    }
}]

messages = [{"role": "user", "content": "What's the weather in Austin, TX?"}]

# Turn 1: Claude decides to call the tool
response = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    tools=tools,
    messages=messages,
)
tool_block = next(b for b in response.content if b.type == "tool_use")
# tool_block.name == "get_weather", tool_block.input == {"location": "Austin, TX"}

# Run your function
def get_weather(location): return "27°C, partly cloudy"
result = get_weather(**tool_block.input)

# Turn 2: send the result back
messages += [
    {"role": "assistant", "content": response.content},
    {"role": "user", "content": [{
        "type": "tool_result",
        "tool_use_id": tool_block.id,
        "content": result
    }]},
]
final = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    tools=tools,
    messages=messages,
)
print(next(b.text for b in final.content if b.type == "text"))
# "The weather in Austin, TX is currently 27°C and partly cloudy."

Things worth knowing

• When does Claude use the tool? By default (tool_choice: {type: "auto"}), Claude decides. Set tool_choice: {type: "any"} to force at least one tool call, or {type: "tool", name: "get_weather"} to force a specific one.
• Strict mode: Add "strict": true to your tool definition. Claude's input will exactly match your schema — no extra fields, no missing required fields. Always pair with "additionalProperties": false.
• Skip the boilerplate: The Tool Runner in the Python and JS SDKs handles the two-turn loop automatically — register your function and the SDK calls it and sends the result.
• Server tools: Anthropic also provides hosted tools (web search, code execution, web fetch) that run on Anthropic's infrastructure — no handler code in your app.

Sources: Tool use overview — Claude Platform Docs · Build a tool-using agent — Claude Platform Docs · Strict tool use — Claude Platform Docs · Claude Function Calling Made Dead Simple — YouTube