Agent SDK
An SDK provided by Anthropic that allows developers to invoke the Claude Code runtime programmatically. It supports tool-call loops, context persistence, and multi-step reasoning coordination, enabling the embedding of Claude Agent capabilities into custom systems and automated workflows.
Core Facts
The core interface of the Agent SDK is a query() function; passing in a prompt and configuration starts the agent loop, and key configuration items include settingSources, allowedTools and systemPrompt.
The free credit can be used for the Claude API, Agent SDK, Managed Agent and Playground.
OpenAI's Agent SDK takes a more engineering-oriented approach, natively supporting agent-to-agent hand-off mechanisms and built-in safety guardrails.
Timeline (last 90 days)
The Cursor Agent SDK's official starter example is a PR review agent that can read code changes and leave comments.
Using the Cursor Agent SDK involves a certain TypeScript learning curve, and people without programming experience will encounter the barriers of text files and TypeScript.
The Cursor Agent SDK provides a security safety net by default: keys are not left exposed in code, and the agent runs in an isolated environment without overstepping into things it should not touch.
The Cursor Agent SDK has a built-in Evals feature that lets developers run a set of tests each time they modify instructions or add tools, to determine whether the agent's performance has improved or worsened.
The Agent SDK is now open to everyone.
The Agent SDK examples mainly come from Cursor itself, and there are currently no independent third-party tests.
The related introduction does not mention any pricing information.
PGA Tour is already using the Agent SDK.
The Agent SDK supports running the agent as a single Node process in a container for self-hosted deployment.
The Agent SDK provides the `agent sdk deploy` command, which automatically builds the agent, assigns a URL, and hosts it to run on Cursor's side.
40 more timeline events
All Facts (20)
The core interface of the Agent SDK is a query() function; passing in a prompt and configuration starts the agent loop, and key configuration items include settingSources, allowedTools and systemPrompt.
65%VerifiedThe free credit can be used for the Claude API, Agent SDK, Managed Agent and Playground.
65%VerifiedOpenAI's Agent SDK takes a more engineering-oriented approach, natively supporting agent-to-agent hand-off mechanisms and built-in safety guardrails.
65%VerifiedAgent SDK is a development toolkit written in TypeScript and Python, and is the core component powering the Claude Code backend
65%UnverifiedAnthropic的Agent SDK支持定义智能体的行动空间和终止条件，内置安全约束和行为边界
85%UnverifiedOpenHands的核心架构基于Agent SDK引擎，利用ReAct框架和Function Calling机制实现代理推理和执行
85%Unverified智能体任务的Token消耗量通常是普通对话的数十倍
75%UnverifiedThe Cursor Agent SDK allows users to define an agent as a folder containing Markdown-format prompts and TypeScript tools, and to review, test, and version-control it like code.
60%UnverifiedClaude Opus 5.5's ecosystem includes Claude Code, Agent SDK, Managed Agents, and it can be called on Bedrock, Google Cloud, and Microsoft Foundry.
60%UnverifiedThird-party applications such as Conductor and OpenCloud are developed based on the Claude Agent SDK
60%UnverifiedThe Cursor Agent SDK's official starter example is a PR review agent that can read code changes and leave comments.
50%UnverifiedThe Cursor Agent SDK provides a security safety net by default: keys are not left exposed in code, and the agent runs in an isolated environment without overstepping into things it should not touch.
50%UnverifiedThe Cursor Agent SDK has a built-in Evals feature that lets developers run a set of tests each time they modify instructions or add tools, to determine whether the agent's performance has improved or worsened.
50%UnverifiedUsing the Cursor Agent SDK involves a certain TypeScript learning curve, and people without programming experience will encounter the barriers of text files and TypeScript.
50%UnverifiedThe Agent SDK supports running the agent as a single Node process in a container for self-hosted deployment.
50%UnverifiedThe Agent SDK provides the `agent sdk deploy` command, which automatically builds the agent, assigns a URL, and hosts it to run on Cursor's side.
50%UnverifiedPGA Tour is already using the Agent SDK.
50%UnverifiedThe related introduction does not mention any pricing information.
50%UnverifiedThe Agent SDK examples mainly come from Cursor itself, and there are currently no independent third-party tests.
50%UnverifiedThe Agent SDK supports mounting guardrails separately at the input, output, or tool level.
50%