Anthropic

使用者入門

將必要的 NuGet 套件新增至您的專案。

dotnet add package Microsoft.Agents.AI.Anthropic --prerelease

如果你正在使用 Microsoft Foundry，也請新增：

dotnet add package Anthropic.Foundry --prerelease

Configuration

環境變數

設定 Anthropic 驗證所需的環境變數：

# Required for Anthropic API access
$env:ANTHROPIC_API_KEY="your-anthropic-api-key"
$env:ANTHROPIC_CHAT_MODEL_NAME="claude-haiku-4-5"  # or your preferred model

您可以從 Anthropic 控制台取得 API 金鑰。

適用於使用 API 金鑰的 Microsoft Foundry

$env:ANTHROPIC_RESOURCE="your-foundry-resource-name"  # Subdomain before .services.ai.azure.com
$env:ANTHROPIC_API_KEY="your-anthropic-api-key"
$env:ANTHROPIC_CHAT_MODEL_NAME="claude-haiku-4-5"

適用於使用 Azure CLI 的 Microsoft Foundry

$env:ANTHROPIC_RESOURCE="your-foundry-resource-name"  # Subdomain before .services.ai.azure.com
$env:ANTHROPIC_CHAT_MODEL_NAME="claude-haiku-4-5"

創建人類代理

基本代理程式建立 (Anthropic Public API)

使用公開 API 建立 Anthropic 代理最簡單的方法：

var apiKey = Environment.GetEnvironmentVariable("ANTHROPIC_API_KEY");
var deploymentName = Environment.GetEnvironmentVariable("ANTHROPIC_CHAT_MODEL_NAME") ?? "claude-haiku-4-5";

AnthropicClient client = new() { ApiKey = apiKey };

AIAgent agent = client.AsAIAgent(
    model: deploymentName,
    name: "HelpfulAssistant",
    instructions: "You are a helpful assistant.");

// Invoke the agent and output the text result.
Console.WriteLine(await agent.RunAsync("Hello, how can you help me?"));

在 Foundry 上使用 Anthropic

在 Microsoft Foundry 上設定好 Anthropic 後，你可以搭配 API 金鑰驗證一起使用：

API 金鑰認證

var resource = Environment.GetEnvironmentVariable("ANTHROPIC_RESOURCE");
var apiKey = Environment.GetEnvironmentVariable("ANTHROPIC_API_KEY");
var deploymentName = Environment.GetEnvironmentVariable("ANTHROPIC_CHAT_MODEL_NAME") ?? "claude-haiku-4-5";

AnthropicClient client = new AnthropicFoundryClient(
    new AnthropicFoundryApiKeyCredentials(apiKey, resource));

AIAgent agent = client.AsAIAgent(
    model: deploymentName,
    name: "FoundryAgent",
    instructions: "You are a helpful assistant using Anthropic on Microsoft Foundry.");

Console.WriteLine(await agent.RunAsync("How do I use Anthropic on Foundry?"));

Azure 認證驗證

對於偏好使用 Azure 憑證的環境：

var resource = Environment.GetEnvironmentVariable("ANTHROPIC_RESOURCE");
var deploymentName = Environment.GetEnvironmentVariable("ANTHROPIC_CHAT_MODEL_NAME") ?? "claude-haiku-4-5";

AnthropicClient client = new AnthropicFoundryClient(
    new AnthropicFoundryIdentityTokenCredentials(
        new DefaultAzureCredential(),
        resource,
        ["https://ai.azure.com/.default"]));

AIAgent agent = client.AsAIAgent(
    model: deploymentName,
    name: "FoundryAgent",
    instructions: "You are a helpful assistant using Anthropic on Microsoft Foundry.");

Console.WriteLine(await agent.RunAsync("How do I use Anthropic on Foundry?"));

Tools

延展性思考

透過原始訊息表示法來設定 Anthropic 推理，並從一般或串流回應中取用 TextReasoningContent。

var apiKey = Environment.GetEnvironmentVariable("ANTHROPIC_API_KEY") ?? throw new InvalidOperationException("ANTHROPIC_API_KEY is not set.");
var model = Environment.GetEnvironmentVariable("ANTHROPIC_CHAT_MODEL_NAME") ?? "claude-haiku-4-5";
var maxTokens = 4096;
var thinkingTokens = 2048;

var agent = new AnthropicClient(new ClientOptions { ApiKey = apiKey })
    .AsAIAgent(
        model: model,
        clientFactory: (chatClient) => chatClient
            .AsBuilder()
            .ConfigureOptions(
                options => options.RawRepresentationFactory = (_) => new MessageCreateParams()
                {
                    Model = options.ModelId ?? model,
                    MaxTokens = options.MaxOutputTokens ?? maxTokens,
                    Messages = [],
                    Thinking = new ThinkingConfigParam(new ThinkingConfigEnabled(budgetTokens: thinkingTokens))
                })
            .Build());

Console.WriteLine("1. Non-streaming:");
var response = await agent.RunAsync("Solve this problem step by step: If a train travels 60 miles per hour and needs to cover 180 miles, how long will the journey take? Show your reasoning.");

Console.WriteLine("#### Start Thinking ####");
Console.WriteLine($"\e[92m{string.Join("\n", response.Messages.SelectMany(m => m.Contents.OfType<TextReasoningContent>().Select(c => c.Text)))}\e[0m");
Console.WriteLine("#### End Thinking ####");

Console.WriteLine("\n#### Final Answer ####");
Console.WriteLine(response.Text);

Console.WriteLine("Token usage:");
Console.WriteLine($"Input: {response.Usage?.InputTokenCount}, Output: {response.Usage?.OutputTokenCount}, {string.Join(", ", response.Usage?.AdditionalCounts ?? [])}");
Console.WriteLine();

Console.WriteLine("2. Streaming");
await foreach (var update in agent.RunStreamingAsync("Explain the theory of relativity in simple terms."))
{
    foreach (var item in update.Contents)
    {
        if (item is TextReasoningContent reasoningContent)
        {
            Console.WriteLine($"\e[92m{reasoningContent.Text}\e[0m");
        }
        else if (item is TextContent textContent)
        {
            Console.WriteLine(textContent.Text);
        }
    }
}

人類技能

由 Anthropic 管理的技能可以透過託管的程式碼執行環境建立檔案。 範例列出可用技能、設定 PowerPoint 技能，並下載產生的檔案。

using Microsoft.Extensions.AI;

string apiKey = Environment.GetEnvironmentVariable("ANTHROPIC_API_KEY") ?? throw new InvalidOperationException("ANTHROPIC_API_KEY is not set.");
// Skills require Claude 4.5 models (Sonnet 4.5, Haiku 4.5, or Opus 4.5)
string model = Environment.GetEnvironmentVariable("ANTHROPIC_CHAT_MODEL_NAME") ?? "claude-sonnet-4-5-20250929";

// Create the Anthropic client
AnthropicClient anthropicClient = new() { ApiKey = apiKey };

// List available Anthropic-managed skills (optional - API may not be available in all regions)
Console.WriteLine("Available Anthropic-managed skills:");
try
{
    SkillListPage skills = await anthropicClient.Beta.Skills.List(
        new SkillListParams { Source = "anthropic", Betas = [AnthropicBeta.Skills2025_10_02] });

    foreach (var skill in skills.Items)
    {
        Console.WriteLine($"  {skill.Source}: {skill.ID} (version: {skill.LatestVersionID})");
    }
}
catch (Exception ex)
{
    Console.WriteLine($"  (Skills listing not available: {ex.Message})");
}

Console.WriteLine();

// Define the pptx skill - the SDK handles all beta flags and container configuration automatically
// when using AsAITool(), so no manual RawRepresentationFactory configuration is needed.
BetaSkillParams pptxSkill = new()
{
    Type = BetaSkillParamsType.Anthropic,
    SkillID = "pptx",
    Version = "latest"
};

// Create an agent with the pptx skill enabled.
// Skills require extended thinking and higher max tokens for complex file generation.
// The SDK's AsAITool() handles beta flags and container config automatically.
ChatClientAgent agent = anthropicClient.Beta.AsAIAgent(
    model: model,
    instructions: "You are a helpful agent for creating PowerPoint presentations.",
    tools: [pptxSkill.AsAITool()],
    clientFactory: (chatClient) => chatClient
        .AsBuilder()
        .ConfigureOptions(options =>
        {
            options.RawRepresentationFactory = (_) => new MessageCreateParams()
            {
                Model = model,
                MaxTokens = 20000,
                Messages = [],
                Thinking = new BetaThinkingConfigParam(
                    new BetaThinkingConfigEnabled(budgetTokens: 10000))
            };
        })
        .Build());

Console.WriteLine("Creating a presentation about renewable energy...\n");

}

// Collect generated files from CodeInterpreterToolResultContent outputs
List<HostedFileContent> hostedFiles = response.Messages
    .SelectMany(m => m.Contents.OfType<CodeInterpreterToolResultContent>())
    .Where(c => c.Outputs is not null)
    .SelectMany(c => c.Outputs!.OfType<HostedFileContent>())
    .ToList();

if (hostedFiles.Count > 0)
{
    Console.WriteLine("\n#### Generated Files ####");
    foreach (HostedFileContent file in hostedFiles)
    {
        Console.WriteLine($"  FileId: {file.FileId}");

        // Download the file using the Anthropic Files API
        using HttpResponse fileResponse = await anthropicClient.Beta.Files.Download(
            file.FileId,
            new FileDownloadParams { Betas = ["files-api-2025-04-14"] });

        // Save the file to disk
        string fileName = $"presentation_{file.FileId.Substring(0, 8)}.pptx";
        using FileStream fileStream = File.Create(fileName);
        Stream contentStream = await fileResponse.ReadAsStream();
        await contentStream.CopyToAsync(fileStream);

使用代理程式

代理程式是標準 AIAgent ，支援所有標準代理程式作業。

請參閱 代理程式入門教學課程 ，以取得如何執行代理程式並與代理程式互動的詳細資訊。

Prerequisites

安裝 Microsoft Agent Framework Anthropic 套件。

pip install agent-framework-anthropic --pre

Configuration

環境變數

設定 Anthropic 驗證所需的環境變數：

# Required for Anthropic API access
ANTHROPIC_API_KEY="your-anthropic-api-key"
ANTHROPIC_CHAT_MODEL="claude-sonnet-4-5-20250929"  # or your preferred model

# Optional: override the Anthropic API endpoint (e.g. for Foundry-compatible deployments)
ANTHROPIC_BASE_URL="https://your-custom-endpoint.com"

您也可以在專案根目錄中使用一個 .env 檔案：

ANTHROPIC_API_KEY=your-anthropic-api-key
ANTHROPIC_CHAT_MODEL=claude-sonnet-4-5-20250929
# ANTHROPIC_BASE_URL=https://your-custom-endpoint.com  # optional

您可以從 Anthropic 控制台取得 API 金鑰。

使用者入門

從代理程式架構匯入所需的類別：

import asyncio
from agent_framework import Agent
from agent_framework.anthropic import AnthropicClient

創建人類代理

基本代理程式創建

創建一個 Anthropic 智能體的最簡單方法：

from agent_framework import Agent

async def basic_example():
    # Create an agent using Anthropic
    agent = Agent(
        client=AnthropicClient(),
        name="HelpfulAssistant",
        instructions="You are a helpful assistant.",
    )

    result = await agent.run("Hello, how can you help me?")
    print(result.text)

使用明確配置

您可以提供明確的設定，而不是依賴環境變數：

from agent_framework import Agent

async def explicit_config_example():
    agent = Agent(
        client=AnthropicClient(
            model="claude-sonnet-4-5-20250929",
            api_key="your-api-key-here",
        ),
        name="HelpfulAssistant",
        instructions="You are a helpful assistant.",
    )

    result = await agent.run("What can you do?")
    print(result.text)

Anthropic SDK 1.x 不支援 temperature、 top_p或 top_k 選項。 代理框架會忽略這些選項，如果你提供這些選項，便會記錄警告。

使用自訂基礎網址

將 base_url 直接傳遞至 AnthropicClient，以指向任何與 Anthropic 相容的端點，例如 Foundry 託管的部署。 這樣你就能保留相同的 AnthropicClient 程式碼，只修改端點，而不必切換到 AnthropicFoundryClient：

from agent_framework import Agent

async def custom_base_url_example():
    agent = Agent(
        client=AnthropicClient(
            model="claude-haiku-4-5",
            api_key="your-api-key-here",
            base_url="https://your-foundry-resource.services.ai.azure.com/models/anthropic",
        ),
        name="HelpfulAssistant",
        instructions="You are a helpful assistant.",
    )

    result = await agent.run("What can you do?")
    print(result.text)

base_url 當未明確傳遞時，會回退到 ANTHROPIC_BASE_URL 環境變數。

在 Foundry 上使用 Anthropic

在 Foundry 上設定好 Anthropic 後，請確保你設定了以下環境變數：

ANTHROPIC_FOUNDRY_API_KEY="your-foundry-api-key"
ANTHROPIC_FOUNDRY_RESOURCE="your-foundry-resource-name"
ANTHROPIC_CHAT_MODEL="claude-haiku-4-5"

接著建立代理人如下：

from agent_framework import Agent
from agent_framework.anthropic import AnthropicFoundryClient

async def foundry_example():
    agent = Agent(
        client=AnthropicFoundryClient(),
        name="FoundryAgent",
        instructions="You are a helpful assistant using Anthropic on Foundry.",
    )

    result = await agent.run("How do I use Anthropic on Foundry?")
    print(result.text)

在 Amazon Bedrock 上使用 Anthropic

AnthropicBedrockClient 透過 Amazon Bedrock 處理 Claude 模型推論。

AWS_ACCESS_KEY_ID="<access-key>"
AWS_SECRET_ACCESS_KEY="<secret-key>"
AWS_REGION="us-east-1"
# Optional:
AWS_PROFILE="<profile>"
AWS_SESSION_TOKEN="<session-token>"
ANTHROPIC_BEDROCK_BASE_URL="<custom-endpoint>"
ANTHROPIC_CHAT_MODEL="anthropic.claude-3-5-sonnet-20241022-v2:0"

目前尚無可執行的代理框架範例公開。AnthropicBedrockClient

在 Google Vertex AI 上使用 Anthropic

AnthropicVertexClient 透過 Google Vertex AI 路由 Claude 模型推論。

CLOUD_ML_REGION="us-east5"
ANTHROPIC_VERTEX_PROJECT_ID="<google-cloud-project>"
ANTHROPIC_CHAT_MODEL="claude-sonnet-4@20250514"
# Optional:
ANTHROPIC_VERTEX_BASE_URL="<custom-endpoint>"

目前尚無可執行的代理框架範例公開。AnthropicVertexClient

處理請求失敗情況

無論是串流或非串流執行，Anthropic SDK 的請求失敗都會透過代理框架例外揭露。 驗證與權限失敗會發生 ChatClientInvalidAuthException，其他 HTTP 4xx 失敗也會發生 ChatClientInvalidRequestException，所有其他提供者失敗也會產生 ChatClientException。

如果 Anthropic 在本地函式呼叫開始後使回應失效，Agent Framework 會引發 ResponseInvalidatedException，且不會執行或保存已失效的呼叫。 串流消費者在例外發生時可能已經收到部分更新;丟棄這些更新，而不是把它們當作已完成的回覆來處理。

當相同的錯誤處理應套用於不同聊天提供者時，請擷取ChatClientException。

Tools

除了標準函式工具支援外，AnthropicClient 還會公開託管的 Anthropic 工具工廠。 使用client.get_*_tool(...)建立工具，並透過tools=在Agent(...)上傳遞它。

如需更豐富的範例 — 結合託管 MCP、Web 搜尋、延伸思考和 Anthropic 技能 — 請參閱下方的託管工具。

代理功能

from typing import Annotated

def get_weather(
    location: Annotated[str, "The location to get the weather for."],
) -> str:
    """Get the weather for a given location."""
    conditions = ["sunny", "cloudy", "rainy", "stormy"]
    return f"The weather in {location} is {conditions[randint(0, 3)]} with a high of {randint(10, 30)}°C."

from agent_framework import Agent

async def tools_example():
    agent = Agent(
        client=AnthropicClient(),
        name="WeatherAgent",
        instructions="You are a helpful weather assistant.",
        tools=get_weather,  # Add tools to the agent
    )

    result = await agent.run("What's the weather like in Seattle?")
    print(result.text)

串流回應

取得即時產生的回應，以提供更好的使用者體驗：

from agent_framework import Agent

async def streaming_example():
    agent = Agent(
        client=AnthropicClient(),
        name="WeatherAgent",
        instructions="You are a helpful weather agent.",
        tools=get_weather,
    )

    query = "What's the weather like in Portland and in Paris?"
    print(f"User: {query}")
    print("Agent: ", end="", flush=True)
    async for chunk in agent.run(query, stream=True):
        if chunk.text:
            print(chunk.text, end="", flush=True)
    print()

託管工具

Anthropic 代理程式支援託管工具，例如 Web 搜尋、MCP（模型上下文協定）和程式碼執行：

from agent_framework import Agent
from agent_framework.anthropic import AnthropicClient

async def hosted_tools_example():
    client = AnthropicClient()
    agent = Agent(
        client=client,
        name="DocsAgent",
        instructions="You are a helpful agent for both Microsoft docs questions and general questions.",
        tools=[
            client.get_mcp_tool(
                name="Microsoft Learn MCP",
                url="https://learn.microsoft.com/api/mcp",
            ),
            client.get_web_search_tool(),
        ],
        default_options={"max_tokens": 20000},
    )

    result = await agent.run("Can you compare Python decorators with C# attributes?")
    print(result.text)

延伸思考（推理）

Anthropic 透過該 thinking 功能支援擴展思考能力，該功能允許模型展示其推理過程：

from agent_framework import Agent
from agent_framework.anthropic import AnthropicClient

async def thinking_example():
    client = AnthropicClient()
    agent = Agent(
        client=client,
        name="DocsAgent",
        instructions="You are a helpful agent.",
        tools=[client.get_web_search_tool()],
        default_options={
            "max_tokens": 20000,
            "thinking": {"type": "enabled", "budget_tokens": 10000}
        },
    )

    query = "Can you compare Python decorators with C# attributes?"
    print(f"User: {query}")
    print("Agent: ", end="", flush=True)

    async for chunk in agent.run(query, stream=True):
        for content in chunk.contents:
            if content.type == "text_reasoning":
                # Display thinking in a different color
                print(f"\033[32m{content.text}\033[0m", end="", flush=True)
            if content.type == "usage":
                print(f"\n\033[34m[Usage: {content.usage_details}]\033[0m\n", end="", flush=True)
        if chunk.text:
            print(chunk.text, end="", flush=True)
    print()

人類技能

Anthropic 提供可管理的技能，擴展客服人員的能力，例如製作 PowerPoint 簡報。 技能需要程式碼解譯工具才能運作：

from typing import cast

from agent_framework import Agent, Content
from agent_framework.anthropic import AnthropicClient
from anthropic import AsyncAnthropic

async def skills_example():
    client = AnthropicClient()
    anthropic_client = cast(AsyncAnthropic, client.anthropic_client)

    # Create an agent with the pptx skill enabled
    # Skills require the Code Interpreter tool
    agent = Agent(
        client=client,
        name="PresentationAgent",
        instructions="You are a helpful agent for creating PowerPoint presentations.",
        tools=client.get_code_interpreter_tool(),
        default_options={
            "max_tokens": 20000,
            "thinking": {"type": "enabled", "budget_tokens": 10000},
            "container": {
                "skills": [{"type": "anthropic", "skill_id": "pptx", "version": "latest"}]
            },
        },
    )

    query = "Create a presentation about renewable energy with 5 slides"
    print(f"User: {query}")
    print("Agent: ", end="", flush=True)

    files: list[Content] = []
    async for chunk in agent.run(query, stream=True):
        for content in chunk.contents:
            match content.type:
                case "text":
                    print(content.text, end="", flush=True)
                case "text_reasoning":
                    print(f"\033[32m{content.text}\033[0m", end="", flush=True)
                case "hosted_file":
                    # Catch generated files
                    files.append(content)

    print("\n")

    # Download generated files
    if files:
        print("Generated files:")
        for idx, file in enumerate(files):
            if file.file_id is None:
                continue
            file_content = await anthropic_client.files.download(
                file_id=file.file_id
            )
            filename = f"presentation-{idx}.pptx"
            with open(filename, "wb") as f:
                await file_content.write_to_file(f.name)
            print(f"File {idx}: {filename} saved to disk.")

完整範例

# Copyright (c) Microsoft. All rights reserved.

import asyncio
from random import randint
from typing import Annotated

from agent_framework import Agent, tool
from agent_framework.anthropic import AnthropicClient

"""
Anthropic Chat Agent Example

This sample demonstrates using Anthropic with an agent and a single custom tool.
"""


# NOTE: approval_mode="never_require" is for sample brevity. Use "always_require" in production; see samples/02-agents/tools/function_tool_with_approval.py and samples/02-agents/tools/function_tool_with_approval_and_sessions.py.
@tool(approval_mode="never_require")
def get_weather(
    location: Annotated[str, "The location to get the weather for."],
) -> str:
    """Get the weather for a given location."""
    conditions = ["sunny", "cloudy", "rainy", "stormy"]
    return f"The weather in {location} is {conditions[randint(0, 3)]} with a high of {randint(10, 30)}°C."


async def non_streaming_example() -> None:
    """Example of non-streaming response (get the complete result at once)."""
    print("=== Non-streaming Response Example ===")

    agent = Agent(
        client=AnthropicClient(),
        name="WeatherAgent",
        instructions="You are a helpful weather agent.",
        tools=get_weather,
    )

    query = "What's the weather like in Seattle?"
    print(f"User: {query}")
    result = await agent.run(query)
    print(f"Result: {result}\n")


async def streaming_example() -> None:
    """Example of streaming response (get results as they are generated)."""
    print("=== Streaming Response Example ===")

    agent = Agent(
        client=AnthropicClient(),
        name="WeatherAgent",
        instructions="You are a helpful weather agent.",
        tools=get_weather,
    )

    query = "What's the weather like in Portland and in Paris?"
    print(f"User: {query}")
    print("Agent: ", end="", flush=True)
    async for chunk in agent.run(query, stream=True):
        if chunk.text:
            print(chunk.text, end="", flush=True)
    print("\n")


async def main() -> None:
    print("=== Anthropic Example ===")

    await streaming_example()
    await non_streaming_example()


if __name__ == "__main__":
    asyncio.run(main())

使用代理程式

代理程式是標準 Agent ，支援所有標準代理程式作業。

請參閱 代理程式入門教學課程 ，以取得如何執行代理程式並與代理程式互動的詳細資訊。

Anthropic

該anthropicprovider套件使用 Anthropic API 建立代理。

Installation

go get github.com/microsoft/agent-framework-go

建立 Anthropic 代理程式

import (
    "github.com/microsoft/agent-framework-go/agent"
    "github.com/microsoft/agent-framework-go/provider/anthropicprovider"

    "github.com/anthropics/anthropic-sdk-go"
)

a := anthropicprovider.NewAgent(
    anthropic.NewClient(), // uses ANTHROPIC_API_KEY env var
    anthropicprovider.AgentConfig{
        Model: "claude-sonnet-4-5",
        Instructions: "You are a helpful assistant.",
        Config: agent.Config{
            Name:         "ClaudeAgent",
        },
    },
)

resp, err := a.RunText(ctx, "Tell me a joke.").Collect()

自訂選項

使用 anthropicprovider.MessageNewParams 傳遞 Anthropic 專屬參數：

resp, err := a.RunText(ctx, "Hello!",
    anthropicprovider.MessageNewParams(anthropic.MessageNewParams{
        MaxTokens: 500,
    }),
).Collect()