Claude Agent SDK
Anthropic's official SDK for production-grade Claude agents
- Best for
- Teams building production agents on top of Claude
- Pricing
- SDK is free, you pay for Claude API usage
- Stack layer
- AI Stack
Live preview of docs.anthropic.com/en/api/agent-sdk-overview
Overview
Claude Agent SDK is the layer Anthropic ships for teams that want to build agents on Claude without rebuilding the orchestration loop themselves. It handles tool calls, multi-turn reasoning, subagents, MCP server wiring, and computer use behind one consistent API across Python and TypeScript.
For AY Automate, this is the default starting point whenever a client asks for a Claude agent that does real work — research, data extraction, ticket triage, browser automation. Skipping the SDK usually means rewriting state, retries, and tool-call parsing from scratch.
The SDK ships in lockstep with Claude model releases, so new features (thinking, computer use, memory) land here first.
Key Features
Tool Use
Define tools as functions; the SDK handles the loop
Subagents
Spawn child agents with scoped context for sub-tasks
MCP Support
First-class Model Context Protocol server integration
Computer Use
Drive a sandboxed desktop directly from the agent
Memory
Long-running context that persists across sessions
TypeScript + Python
Same API surface across both ecosystems
Why We Recommend Claude Agent SDK
It's the shortest path from idea to a working Claude agent. The state model is sound, tool calls are predictable, and you get every Claude capability the day it ships.