Can I use this SDK with non-Anthropic models like GPT-4 or Gemini?

No, the Claude Agent SDK is specifically optimized for and locked to Anthropic's Claude models. It supports Claude via the direct Anthropic API, Amazon Bedrock, Google Vertex AI, or Azure AI Foundry.

Does it support Model Context Protocol (MCP)?

Yes, the SDK allows you to connect to external MCP servers via stdio, SSE, or HTTP, and also lets you create in-process MCP tools using Zod schemas for structured data validation.

What is the Claude Agent SDK?

It is a TypeScript library that allows you to run Claude's agentic loop programmatically. It provides access to the same tools as the Claude Code CLI—like Read, Write, and Bash—but through a structured API.

How do I handle session persistence with the SDK?

You can capture the session_id from the initial system message and pass it as a 'resume' option in subsequent queries to maintain context across different execution blocks.

How do permissions work in the SDK?

You can use built-in modes like 'acceptEdits' or 'plan', or implement a custom 'canUseTool' callback to approve or deny specific tool executions based on the command or file path.

Claude Agent SDK (TypeScript)

Name: Claude Agent SDK (TypeScript)
Author: AnExiledDev

byAnExiledDev

•

API開発

Streamlines the development of custom AI agents using the official Anthropic Claude Agent SDK for TypeScript integration.

This skill provides a comprehensive implementation guide and pattern library for the @anthropic-ai/claude-agent-sdk. It enables developers to integrate Claude's core agentic capabilities—including file manipulation, bash execution, and context management—directly into TypeScript applications. By offering structured patterns for the query function, custom permission hooks, and MCP tool integration, it allows you to transform Claude into a programmable engine for custom devtools, automated workflows, and headless AI assistants.

主な機能

016 GitHub stars

02Session management for capturing, resuming, and forking agentic execution states

03In-process and external MCP (Model Context Protocol) tool integration via Zod schemas

04Implementation patterns for the core query() function and async message streaming

05Advanced permission management using canUseTool callbacks and granular permission modes

06Configuration guidance for multi-cloud providers including AWS Bedrock, Google Vertex AI, and Azure

ユースケース

01Developing domain-specific agents that utilize custom internal APIs via MCP tool servers

02Building custom headless AI coding assistants for specific enterprise environments

03Creating automated security audit agents with restricted, read-only tool access

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add anexileddev/codeforge claude-agent-sdk

For use in Claude.ai and ChatGPT

Download Skill