How does session management work in the SDK?

The SDK provides session IDs that allow you to resume previous conversations or fork them into new paths, preserving the context and history of the agent's interactions.

What is the benefit of using subagents?

Subagents allow you to break down complex tasks by delegating specific parts of a workflow to specialized agent definitions with their own unique prompts, tools, and model configurations.

Can I restrict which tools the agent can use?

Yes, you can use the allowedTools and disallowedTools options to whitelist or blacklist specific capabilities, or implement a canUseTool callback for fine-grained, logic-based permissions.

What is the Claude Agent SDK?

The Claude Agent SDK is a TypeScript library that allows developers to programmatically interact with Claude Code to build autonomous agents, manage custom tools, and orchestrate complex multi-step workflows.

Does this SDK require the Claude Code CLI?

Yes, the Claude Code CLI is a prerequisite. You must have @anthropic-ai/claude-code installed on your system for the SDK to orchestrate tools and execute commands.

Claude Agent SDK

Name: Claude Agent SDK
Author: secondsky

bysecondsky

•

개발자 도구

Orchestrates autonomous AI agents and multi-step workflows through a programmatic interface for Claude Code's tool-using capabilities.

소개

The Claude Agent SDK provides a robust TypeScript framework for building sophisticated agentic systems powered by Anthropic's Claude. It enables developers to move beyond simple chat interfaces into autonomous operations, offering programmatic access to tool orchestration, Model Context Protocol (MCP) servers, and multi-agent systems. With built-in support for session persistence, conversational forking, and granular permission controls, this SDK is designed for creating production-ready coding assistants, automated SRE tools, and specialized dev-ops workflows that require reliable, long-running agentic loops.

주요 기능

21 GitHub stars
Programmatic query API for building autonomous agentic loops
Multi-agent orchestration with specialized subagent definitions
Model Context Protocol (MCP) integration for custom tool extension
Advanced session management with persistence and forking support
Granular security through custom tool-use permission callbacks

사용 사례

Building autonomous CI/CD agents for automated code repair and refactoring
Creating specialized subagents for security auditing and compliance monitoring
Orchestrating complex software engineering tasks requiring multiple tools and persistent state

소개

주요 기능

21 GitHub stars
Programmatic query API for building autonomous agentic loops
Multi-agent orchestration with specialized subagent definitions
Model Context Protocol (MCP) integration for custom tool extension
Advanced session management with persistence and forking support
Granular security through custom tool-use permission callbacks

사용 사례

Building autonomous CI/CD agents for automated code repair and refactoring
Creating specialized subagents for security auditing and compliance monitoring
Orchestrating complex software engineering tasks requiring multiple tools and persistent state