Claude Agent SDK
The Claude Agent SDK is a software development kit provided by Anthropic to help developers build custom AI agents based on the Claude Code Agent Harness framework. The SDK supports both Python and TypeScript, providing standardized interfaces and toolchains that enable developers to extend and customize AI agent behaviors and capabilities. It is applicable to scenarios such as automated task execution and workflow orchestration.
Core Facts
Claude Agent SDK supports multi-step reasoning, tool calls, and environment interaction, including capabilities such as code writing, file operations, and command execution.
The Claude Agent SDK has built-in tools such as Web Search and Web Fetch, giving the Agent the ability to search the internet and scrape web pages.
Timeline (last 90 days)
一项针对12个主流Agent框架的调查发现，没有任何一个框架在文档中提供以货币计价的工具调用预算控制（C5全部为否）
Pexafy's official documentation lists tested configurations for the OpenAI Agents SDK, Claude Agent SDK, Google ADK, CrewAI, LlamaIndex, Mastra and Strands Agents.
Subagents in the Claude Agent SDK cannot spawn their own subagents, so the architecture should be a flat fan-out rather than a tree structure.
The Claude Agent SDK's PreToolUse and PostToolUse hooks, combined with matchers such as Write|Edit, can block, modify, or log tool calls.
Pink Agentic AI Payments conducted a systematic review of the 12 most popular Agent frameworks and concluded that none of the frameworks documents a currency-denominated tool-call budget control (C5).
Both SDKs define tools using one function paired with one decorator — Claude uses the tool decorator and OpenAI uses the function_tool decorator.
In Claude, guards are called pre_tool_use hooks paired with a deny response; in OpenAI, the counterpart is tool_input_guardrail, which returns a rejection payload.
Claude's sub-agents are defined via options.agents, and the background=False field is used to make the main agent wait for the sub-agent to return before the main agent composes its reply.
Both the Claude Agent SDK and the OpenAI Agents SDK take over the Agent Loop, greatly simplifying the model invocation, tool dispatching, history maintenance, and loop control logic that previously had to be written by hand.
The Claude Agent SDK invokes via query (with two core fields), while the OpenAI Agents SDK uses Runner.run_async with agent and query passed in.
20 more timeline events
All Facts (20)
Claude Agent SDK supports multi-step reasoning, tool calls, and environment interaction, including capabilities such as code writing, file operations, and command execution.
65%VerifiedThe Claude Agent SDK has built-in tools such as Web Search and Web Fetch, giving the Agent the ability to search the internet and scrape web pages.
65%Unverifiedclaude-memory-compiler采用三层自动化流水线架构：会话捕获层（Hooks机制）、知识提取层（Claude Agent SDK）和知识编译层（LLM Compiler）
90%Unverified该方案的所有数据和推理都在本地完成，不会外传到第三方服务器，适合对数据隐私有严格要求的企业场景
90%Unverified通过设置环境变量ANTHROPIC_BASE_URL指向LiteLLM Proxy的本地端口，可以将Claude Agent SDK的请求重定向到本地模型，无需修改SDK源码
90%Unverified2024-2025年，Anthropic Claude Agent SDK、OpenAI Agents SDK、Google ADK等Agent框架集中涌现
85%UnverifiedClaude Agent SDK提供工具注册与调用、多轮对话上下文管理、智能体间消息传递与协调机制以及结构化输出控制等能力
85%Unverified框架能力与推理能力解耦是AI工程领域的重要趋势，Claude Agent SDK提供Agent编排框架，实际推理可由任何兼容模型完成
75%Unverified该方案的整体架构分为三层：Agent Framework层（任务管理和调度）、Claude Agent SDK层（核心Worker）、LiteLLM Proxy层（请求代理转发）
60%UnverifiedThe solution of connecting the Claude Agent SDK to a local large model via LiteLLM Proxy can reduce API call costs from nearly $1 per simple task to almost zero.
60%Unverified一项针对12个主流Agent框架的调查发现，没有任何一个框架在文档中提供以货币计价的工具调用预算控制（C5全部为否）
50%UnverifiedPexafy's official documentation lists tested configurations for the OpenAI Agents SDK, Claude Agent SDK, Google ADK, CrewAI, LlamaIndex, Mastra and Strands Agents.
50%UnverifiedThe Claude Agent SDK's PreToolUse and PostToolUse hooks, combined with matchers such as Write|Edit, can block, modify, or log tool calls.
50%UnverifiedSubagents in the Claude Agent SDK cannot spawn their own subagents, so the architecture should be a flat fan-out rather than a tree structure.
50%UnverifiedPink Agentic AI Payments conducted a systematic review of the 12 most popular Agent frameworks and concluded that none of the frameworks documents a currency-denominated tool-call budget control (C5).
50%UnverifiedTriggering max_turns means the agent has timed out, which is an abnormal rather than a normal termination signal — Claude returns result error and OpenAI throws max turn exceeded, and neither should be treated as a normal stop answer.
50%UnverifiedBoth the Claude Agent SDK and the OpenAI Agents SDK take over the Agent Loop, greatly simplifying the model invocation, tool dispatching, history maintenance, and loop control logic that previously had to be written by hand.
50%UnverifiedThe Claude Agent SDK invokes via query (with two core fields), while the OpenAI Agents SDK uses Runner.run_async with agent and query passed in.
50%UnverifiedBoth SDKs define tools using one function paired with one decorator — Claude uses the tool decorator and OpenAI uses the function_tool decorator.
50%UnverifiedIn Claude, guards are called pre_tool_use hooks paired with a deny response; in OpenAI, the counterpart is tool_input_guardrail, which returns a rejection payload.
50%