How does session forking help in agent development?

Session forking allows you to create a new branch from an existing conversation, enabling the agent to explore alternative code paths or solutions without affecting the original session state.

What is the Model Context Protocol (MCP) support?

The SDK supports connecting to MCP servers via stdio, HTTP, or SSE transports, allowing your agents to use standardized tools for filesystem access, database queries, or external API integrations.

How do subagents work in this SDK?

You can define specialized agents with their own system prompts and toolsets. The main agent can then delegate specific tasks, such as security auditing or UI testing, to these subagents.

What are structured outputs in the Claude Agent SDK?

Structured outputs allow you to define a specific JSON schema or Zod object that Claude's response must strictly follow, ensuring guaranteed validation and type safety for your application logic.

Can I control which tools the agent is allowed to execute?

Yes, the SDK provides a 'canUseTool' callback where you can implement custom logic to allow, deny, or request user confirmation based on the specific tool name and input parameters.

Claude Agent SDK

Name: Claude Agent SDK
Author: jezweb

byjezweb

•

開発者ツール

Builds autonomous AI agents with guaranteed structured JSON outputs and advanced event-driven workflows.

概要

The Claude Agent SDK provides a comprehensive framework for developing sophisticated, production-ready AI agents powered by Claude. It enables developers to implement strict JSON schema validation via structured outputs, manage complex tool permissions, and orchestrate specialized subagents for distinct tasks. With advanced features like session forking for non-destructive experimentation, a robust hooks system for event-driven logic, and seamless Model Context Protocol (MCP) integration, this skill is essential for building reliable coding assistants, automated SRE systems, and security auditors that require type-safe, predictable responses.

主な機能

91 GitHub stars
Comprehensive hooks system for event-driven workflows (PreToolUse, PostToolUse, etc.).
Deep integration with Model Context Protocol (MCP) for external tool and data access.
Structured JSON output validation using Zod or JSON schema for type-safe responses.
Advanced session management with support for forking, resuming, and state persistence.
Granular permission control via custom callbacks to allow, deny, or prompt for tool usage.

ユースケース

Developing autonomous coding agents that require validated, schema-compliant JSON outputs.
Orchestrating multi-agent systems for complex tasks like security audits or automated troubleshooting.
Building interactive CLI tools that utilize MCP servers for filesystem and git history analysis.

概要

主な機能

91 GitHub stars
Comprehensive hooks system for event-driven workflows (PreToolUse, PostToolUse, etc.).
Deep integration with Model Context Protocol (MCP) for external tool and data access.
Structured JSON output validation using Zod or JSON schema for type-safe responses.
Advanced session management with support for forking, resuming, and state persistence.
Granular permission control via custom callbacks to allow, deny, or prompt for tool usage.

ユースケース

Developing autonomous coding agents that require validated, schema-compliant JSON outputs.
Orchestrating multi-agent systems for complex tasks like security audits or automated troubleshooting.
Building interactive CLI tools that utilize MCP servers for filesystem and git history analysis.