How does the SDK ensure agent safety and security?

It provides a permission system where you can define 'allowedTools' and 'disallowedTools', along with a 'strict' permission mode to prevent unauthorized code execution or file access.

Can I use models from providers other than Anthropic directly?

Yes, the SDK supports Anthropic API as the default, but it can be configured to use Claude via Amazon Bedrock, Google Vertex AI, or Microsoft Foundry.

Which programming languages does the Claude Agent SDK support?

The SDK is available for both TypeScript (via npm) and Python (via pip), allowing you to build agents in your preferred ecosystem.

What is the role of MCP in this SDK?

The Model Context Protocol (MCP) allows you to connect your agent to external data sources and services, like Postgres databases or custom APIs, using standardized tool servers.

Claude Agent SDK Guide

Name: Claude Agent SDK Guide
Author: majesticlabs-dev

bymajesticlabs-dev

•

開発者ツール

Builds custom, production-ready AI agents using the same infrastructure and tool ecosystem that powers Claude Code.

概要

The Claude Agent SDK provides a robust framework for developers to build specialized AI agents in TypeScript and Python using Anthropic's powerful models. It offers advanced context management through automatic compaction, a comprehensive tool ecosystem including file operations and bash execution, and granular permission controls. With support for multiple authentication providers like AWS Bedrock and Google Vertex AI, plus extensibility via the Model Context Protocol (MCP), this skill enables the creation of sophisticated agents for SRE diagnostics, security auditing, and automated business workflows.

主な機能

Automated context compaction for long-running agent sessions
Cross-language support for TypeScript and Python SDKs
Model Context Protocol (MCP) support for custom tool integration
Multi-provider authentication for Bedrock, Vertex AI, and Foundry
Granular permission controls for file, bash, and web tools
18 GitHub stars

ユースケース

Building secure code review assistants with restricted file-system access
Creating business agents for automated legal review and document synthesis
Developing SRE agents that analyze logs and suggest infrastructure fixes

概要

主な機能

Automated context compaction for long-running agent sessions
Cross-language support for TypeScript and Python SDKs
Model Context Protocol (MCP) support for custom tool integration
Multi-provider authentication for Bedrock, Vertex AI, and Foundry
Granular permission controls for file, bash, and web tools
18 GitHub stars

ユースケース

Building secure code review assistants with restricted file-system access
Creating business agents for automated legal review and document synthesis
Developing SRE agents that analyze logs and suggest infrastructure fixes