跳转至

Anthropic 集成

Changelog

New in version: 0.7.0

本指南介绍如何将 ToolRegistry 与 Anthropic API（Claude）配合使用。ToolRegistry 可以生成 Anthropic 原生的工具 Schema，并重建 tool_use / tool_result 消息以支持多轮对话。

设置 ToolRegistry

from toolregistry import ToolRegistry

registry = ToolRegistry()

@registry.register
def add(a: float, b: float) -> float:
    """Add two numbers together."""
    return a + b

@registry.register
def subtract(a: float, b: float) -> float:
    """Subtract the second number from the first."""
    return a - b

导出工具 Schema

schemas = registry.get_schemas(api_format="anthropic")

返回 Anthropic 格式的工具定义：

[
  {
    "name": "add",
    "description": "Add two numbers together.",
    "input_schema": {
      "properties": {
        "a": { "title": "A", "type": "number" },
        "b": { "title": "B", "type": "number" }
      },
      "required": ["a", "b"],
      "title": "addParameters",
      "type": "object"
    }
  },
  {
    "name": "subtract",
    "description": "Subtract the second number from the first.",
    "input_schema": {
      "properties": {
        "a": { "title": "A", "type": "number" },
        "b": { "title": "B", "type": "number" }
      },
      "required": ["a", "b"],
      "title": "subtractParameters",
      "type": "object"
    }
  }
]

使用工具 Schema 发送查询

import os
import anthropic

client = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))

messages = [
    {"role": "user", "content": "I have 15 chestnuts. Joe ate 3. How many do I have left?"}
]

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    tools=schemas,
    messages=messages,
)

提取工具调用

Anthropic 以 tool_use 内容块的形式返回工具调用：

tool_calls = [block for block in response.content if block.type == "tool_use"]

tool_use 块示例：

{
  "type": "tool_use",
  "id": "toolu_01A09q90qw90lq917835lq9",
  "name": "subtract",
  "input": {"a": 15, "b": 3}
}

执行工具调用

ToolRegistry 原生支持 Anthropic 的 tool_use 块：

tool_responses = registry.execute_tool_calls(tool_calls)

返回一个字典，键为工具调用 ID，值为执行结果：

{"toolu_01A09q90qw90lq917835lq9": "12"}

将结果反馈给 LLM

以 Anthropic 格式重建对话消息：

assistant_tool_messages = registry.build_tool_call_messages(
    tool_calls, tool_responses, api_format="anthropic"
)

生成 Anthropic 原生的消息结构：

[
  {
    "role": "assistant",
    "content": [
      {
        "type": "tool_use",
        "id": "toolu_01A09q90qw90lq917835lq9",
        "name": "subtract",
        "input": {"a": 15, "b": 3}
      }
    ]
  },
  {
    "role": "user",
    "content": [
      {
        "type": "tool_result",
        "tool_use_id": "toolu_01A09q90qw90lq917835lq9",
        "content": "12"
      }
    ]
  }
]

将消息追加到对话中并获取最终答案：

messages.extend(assistant_tool_messages)

second_response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=messages,
)

print(second_response.content[0].text)

完整 Python 脚本

import json
import os

import anthropic
from toolregistry import ToolRegistry

registry = ToolRegistry()

@registry.register
def add(a: float, b: float) -> float:
    """Add two numbers together."""
    return a + b

@registry.register
def subtract(a: float, b: float) -> float:
    """Subtract the second number from the first."""
    return a - b

client = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))

messages = [
    {"role": "user", "content": "I have 15 chestnuts. Joe ate 3. How many do I have left?"}
]

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    tools=registry.get_schemas(api_format="anthropic"),
    messages=messages,
)

tool_calls = [block for block in response.content if block.type == "tool_use"]

if tool_calls:
    tool_responses = registry.execute_tool_calls(tool_calls)
    print(tool_responses)

    assistant_tool_messages = registry.build_tool_call_messages(
        tool_calls, tool_responses, api_format="anthropic"
    )
    print(json.dumps(assistant_tool_messages, indent=2))

    messages.extend(assistant_tool_messages)
    second_response = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=1024,
        messages=messages,
    )

    print(second_response.content[0].text)

参见