Claude Agent SDK

The Claude Agent SDK is Anthropic's opinionated framework for building agents that use Claude. Built on the same primitives as Claude Code, it handles tool use, multi-turn orchestration, and sub-agents.

What it provides

• Tool-use loop orchestration
• Structured tool definitions via TypeScript/Python
• Sub-agent spawning and delegation
• Built-in support for MCP tool servers
• File system, shell, and web tools out of the box
• Streaming output

When to use

• Building on Claude
• Want opinions on how to structure agents (not a DIY framework)
• Need MCP integration
• Want close behavior parity with how Claude Code works

When it's not the right fit

• Multi-model (GPT, Gemini in the same system), use LangGraph or CrewAI
• Highly custom orchestration logic, build your own on raw API
• Research settings where you want control over every step

Relationship to Claude Code

Claude Code is the end-user CLI. The Agent SDK is the library you'd use to build your own agents with similar capabilities. Think of Claude Code as an agent built with the SDK.

MCP integration

Model Context Protocol is Anthropic's standard for tool servers. The Agent SDK natively supports MCP servers, so you can plug in community-built tool providers without writing glue code.

Getting started

See docs at anthropic.com. Typical starter: npm/pip install the SDK, define tools, define system prompt, launch agent loop. Most production agents fit in a few hundred lines of code.