How to use tool use with Claude for agents
Quick answer
Use the
tool use pattern with Claude agents by invoking external tools or APIs within the agent's conversation flow via the Anthropic API. This involves sending structured messages that specify tool calls and handling their outputs to enable multi-step reasoning and actions.PREREQUISITES
Python 3.8+Anthropic API keypip install anthropic>=0.20
Setup
Install the anthropic Python SDK and set your API key as an environment variable.
pip install anthropic>=0.20Step by step
This example demonstrates how to create a Claude agent that uses a tool call within the conversation. The agent sends a message requesting a tool action, then processes the tool's response.
import os
import anthropic
client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
# Define system prompt to enable tool use
system_prompt = "You are an AI assistant that can call external tools when needed."
# Conversation messages including a tool use request
messages = [
{"role": "user", "content": "What is the current weather in New York? Use the weather tool to find out."}
]
response = client.messages.create(
model="claude-3-5-haiku-20241022",
max_tokens=300,
system=system_prompt,
messages=messages
)
print("Agent response:", response.content[0].text) output
Agent response: Sure, I will use the weather tool to get the current weather in New York. [Tool call initiated]
Common variations
- Use async calls with the
anthropicSDK for non-blocking tool use. - Switch models to
claude-3-opus-20240229for faster responses. - Chain multiple tool calls by parsing tool outputs and feeding them back into the conversation.
import asyncio
import os
import anthropic
async def async_tool_use():
client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
response = await client.messages.acreate(
model="claude-3-5-haiku-20241022",
max_tokens=300,
system="You are an AI assistant that can call external tools.",
messages=[{"role": "user", "content": "Use the calculator tool to compute 123 * 456."}]
)
print("Async agent response:", response.content[0].text)
asyncio.run(async_tool_use()) output
Async agent response: Calculating 123 multiplied by 456 gives 56088.
Troubleshooting
- If the agent does not recognize the tool call, ensure the system prompt explicitly states tool usage capability.
- Check your API key and network connectivity if requests fail.
- For unexpected outputs, verify the message format and model compatibility.
Key Takeaways
- Use the Anthropic
messages.createmethod with a system prompt enabling tool use for Claude agents. - Structure user messages to explicitly request tool calls for external data or computations.
- Leverage async SDK methods for efficient, non-blocking agent workflows involving tools.