Extending Capabilities of Claude Agent SDK in TypeScript

Übersicht

Extend Claude's capabilities beyond basic queries by building richer, stateful, and more integrated agent experiences. You’ll learn to maintain context with long-running clients, connect external MCP servers, wrap your own TypeScript functions as custom tools, and orchestrate specialized sub-agents—turning Claude into a flexible, multi-tool assistant tailored to your application.

Lehrplan

• Unit 1: Maintaining Agent State with Sessions
Observing System Initialization Messages
Capturing Session IDs from Message Streams
Building a Reusable Response Display Helper
Completing Your First Multi-Turn Conversation
• Unit 2: Integrating External MCP Servers
Guiding Your Agent to Use MCP Tools
Fix MCP Tool Naming Convention
Renaming MCP Servers with Descriptive Names
Integrating External MCP Servers
• Unit 3: Creating Custom MCP Tools
Testing Planning Functions Before Creating MCP Tools
Transform Functions into MCP Tools
Connecting Custom MCP Tools to an Agent
Configuring an Agent with Custom Planning Tools
• Unit 4: Building Multi Agent Systems
Creating Your First AgentDefinition
Registering Your First Sub-Agent
Observing Agent Delegation with the Task Tool
Complete the Multi-Agent Code Review System
Integrating MCP Tools into Multi-Agent Systems

---

Fachgebiete

Artificial Intelligence