Skip to main content
Messages API models tool workflows through content blocks rather than tool_calls arrays. Use this page when your app already depends on Anthropic-style tool semantics.

​
Define Tools

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.naga.ac",
    api_key="YOUR_API_KEY",
)

message = client.messages.create(
    model="claude-sonnet-4.5",
    max_tokens=256,
    messages=[
        {
            "role": "user",
            "content": "Check the weather in Prague and tell me if I need a coat.",
        }
    ],
    tools=[
        {
            "name": "lookup_weather",
            "description": "Look up current weather for a city.",
            "input_schema": {
                "type": "object",
                "properties": {
                    "city": {"type": "string"}
                },
                "required": ["city"],
            },
        }
    ],
)

if message.stop_reason == "tool_use":
    tool_use = next(block for block in message.content if block.type == "tool_use")
    print(tool_use.name)
    print(tool_use.input)

​
Assistant Tool Use Block

{
  "type": "tool_use",
  "id": "call_1",
  "name": "lookup_weather",
  "input": {
    "city": "Prague"
  }
}

​
Return Tool Results

Tool results go back in a user message with tool_result blocks. 
{
  "model": "claude-sonnet-4.5",
  "max_tokens": 256,
  "messages": [
    {
      "role": "assistant",
      "content": [
        {
          "type": "tool_use",
          "id": "call_1",
          "name": "lookup_weather",
          "input": {
            "city": "Prague"
          }
        }
      ]
    },
    {
      "role": "user",
      "content": [
        {
          "type": "tool_result",
          "tool_use_id": "call_1",
          "content": "{\"temperature_c\":7,\"raining\":true}"
        }
      ]
    }
  ]
}
Keep the assistant tool_use block in the replayed history before you send the tool_result block. Some reasoning-capable models also return thinking blocks alongside tool_use. If you want continuity across turns, replay those blocks unchanged, including signature, before the later tool_result block. See Messages Thinking Blocks.

​
Streaming Behavior

When streaming is enabled:
  • the tool-use block begins with content_block_start
  • argument deltas arrive as content_block_delta with type: input_json_delta
  • tool-oriented turns usually end with message_delta.stop_reason: "tool_use"

​
Caveats

  • assistant messages can include tool_use blocks but not tool_result blocks
  • user messages can include tool_result blocks but not tool_use blocks
  • tool_choice: {"type":"any"} maps to required tool use semantics internally

​
Common mistakes

  • sending tool_result in an assistant message instead of a user message
  • losing the original tool_use block before sending the result back
  • trying to use Chat Completions or Responses tool payload shapes on this surface