Tool use patterns

Tool Use Patterns

Patterns for defining, orchestrating, and handling Claude tool use. Covers schemas, multi-tool flows, parallel execution, error handling, and result formatting.

When to Use

• User is defining tools for the Claude API
• User is building agentic loops with tool calling
• User needs parallel tool execution
• User is handling tool errors or formatting results
• User is designing multi-step tool workflows

Core Patterns

Tool Definition Schema

Tools are defined with a name, description, and JSON Schema for input_schema. The description is critical -- Claude uses it to decide when to call the tool.

tools = [
    {
        "name": "get_weather",
        "description": "Get current weather for a city. Use this when the user asks about weather, temperature, or forecast for a specific location.",
        "input_schema": {
            "type": "object",
            "properties": {
                "city": {
                    "type": "string",
                    "description": "City name, e.g. 'San Francisco, CA'"
                },
                "units": {
                    "type": "string",
                    "enum": ["celsius", "fahrenheit"],
                    "description": "Temperature units. Default: fahrenheit."
                }
            },
            "required": ["city"]
        }
    }
]

message = client.messages.create(
    model="claude-sonnet-4-6-20250514",
    max_tokens=1024,
    tools=tools,
    messages=[{"role": "user", "content": "What's the weather in Tokyo?"}]
)

Agentic Tool Loop

The core pattern: send a message, check if Claude wants to use tools, execute them, return results, and repeat until Claude produces a final text response.

def run_agent(user_message: str, tools: list, system: str = "") -> str:
    messages = [{"role": "user", "content": user_message}]

    while True:
        response = client.messages.create(
            model="claude-sonnet-4-6-20250514",
            max_tokens=4096,
            system=system,
            tools=tools,
            messages=messages,
        )

        # Collect tool use blocks and text
        tool_results = []
        final_text = ""

        for block in response.content:
            if block.type == "tool_use":
                # Execute the tool
                result = execute_tool(block.name, block.input)
                tool_results.append({
                    "type": "tool_result",
                    "tool_use_id": block.id,
                    "content": json.dumps(result)
                })
            elif block.type == "text":
                final_text = block.text

        # If no tool calls, return the final text
        if response.stop_reason == "end_turn":
            return final_text

        # Append assistant response and tool results
        messages.append({"role": "assistant", "content": response.content})
        messages.append({"role": "user", "content": tool_results})

Parallel Tool Calls

Claude can call multiple tools simultaneously. Handle all tool results in a single user message to maintain correct conversation structure.

# Claude may return multiple tool_use blocks in one response
# Example: user asks "Compare weather in NYC and London"

for block in response.content:
    if block.type == "tool_use":
        # Execute each tool call
        result = execute_tool(block.name, block.input)
        tool_results.append({
            "type": "tool_result",
            "tool_use_id": block.id,  # Must match the specific tool_use block
            "content": json.dumps(result)
        })

# Return ALL results in a single user message
messages.append({"role": "assistant", "content": response.content})
messages.append({"role": "user", "content": tool_results})

Tool Error Handling

Return errors as tool results so Claude can reason about them and recover.

def execute_tool(name: str, inputs: dict) -> dict:
    try:
        if name == "get_weather":
            return get_weather(**inputs)
        elif name == "search_database":
            return search_database(**inputs)
        else:
            return {"error": f"Unknown tool: {name}"}
    except ValueError as e:
        return {"error": f"Invalid input: {str(e)}"}
    except TimeoutError:
        return {"error": "Request timed out. Try again or use a different query."}
    except Exception as e:
        return {"error": f"Tool execution failed: {str(e)}"}

# Alternatively, use the is_error flag for explicit error signaling
tool_results.append({
    "type": "tool_result",
    "tool_use_id": block.id,
    "content": "City not found. Please check the spelling.",
    "is_error": True  # Tells Claude this is an error, not a valid result
})

Forcing Tool Use

Control whether Claude must use a tool, can choose, or must not use tools.

# Force Claude to use a specific tool
message = client.messages.create(
    model="claude-sonnet-4-6-20250514",
    max_tokens=1024,
    tools=tools,
    tool_choice={"type": "tool", "name": "get_weather"},  # Must use this tool
    messages=[{"role": "user", "content": "Tokyo forecast"}]
)

# Force Claude to use any tool (but must pick one)
tool_choice = {"type": "any"}

# Let Claude decide (default behavior)
tool_choice = {"type": "auto"}

Tool Result Formatting

Structure tool results so Claude can reason about them effectively.

# Good: structured, clear result
result = {
    "status": "success",
    "data": {
        "city": "Tokyo",
        "temperature": 22,
        "units": "celsius",
        "conditions": "partly cloudy",
        "humidity": 65
    }
}

# Good: include metadata for pagination or follow-up
result = {
    "status": "success",
    "results": items[:10],
    "total_count": 247,
    "has_more": True,
    "next_cursor": "abc123"
}

Anti-Patterns

• Writing vague tool descriptions that do not explain WHEN to use the tool
• Not including required fields in the input schema
• Returning raw exception tracebacks as tool results (expose internals, confuse Claude)
• Sending tool results with mismatched tool_use_id (causes API error)
• Not handling the case where Claude calls multiple tools in parallel
• Using tool_choice: {"type": "tool"} in a loop (causes infinite tool calling)
• Putting business logic in tool descriptions instead of the system prompt
• Returning excessively large tool results (summarize or paginate instead)

Quick Reference

Field	Purpose
name	Tool identifier, snake_case, max 64 chars
description	When and why to use -- be specific and directive
input_schema	JSON Schema object with properties and required
tool_use_id	Links tool result back to the specific tool call
is_error	Boolean flag signaling error in tool result
tool_choice	auto (default), any (must call one), {"type":"tool","name":"X"}

Conversation structure for tool use:

1. user message -> Claude responds with tool_use blocks
2. assistant message (Claude's response with tool_use) -> user message with tool_result blocks
3. Repeat until stop_reason is end_turn