Apr 17
Loading video player...
Claude Agent SDK Template: https://replit.com/@matt/Claude-Agent-SDK?v=1?utm_source=Matt&utm_medium=youtube&utm_campaign=agent-sdk Replit Integrations: https://replit.com/integrations?utm_source=Matt&utm_medium=youtube&utm_campaign=agent-sdk Skills.sh Framework: https://skills.sh Build autonomous agents that handle repetitive workflows. You'll learn the fundamental architecture behind coding agents like Claude Code and Replit Agent, then apply those concepts to create your own scheduled automation. Key Points: 00:00 Introduction to Claude Agent SDK as Claude Code in library form 01:08 Understanding the agent loop: prompt, tools, observe, repeat 01:38 Multi-agent architectures and orchestrator patterns explained 03:23 Core primitives: query loops, agents, tools, MCP servers, skills, permissions 05:30 Tools vs MCP servers vs skills: when to use each 07:08 Context windows and why specialized sub-agents matter 10:47 Building on Replit: remixing templates and project structure 15:17 Creating a Todoist organizer agent with custom tools 22:06 Debugging OAuth limitations and API integration strategies 28:23 Deploying scheduled automations with Replit publish Social Links: X: https://x.com/mattppal LinkedIn: https://www.linkedin.com/in/matt-palmer #ClaudeAgentSDK #AIAgents #Replit #Automation #VibeCoding #AIProductivity #NoCode #LowCode #Todoist #MCPServers What repetitive task would you automate first with the Claude Agent SDK? Share your ideas in the comments, and let me know if you'd like a deeper dive into multi-agent architectures or MCP server development.