Claude Agent SDK

The Claude Agent SDK is Anthropic's answer to the growing need for a first-party framework that makes it easier to build reliable agentic systems on top of Claude models, rather than wiring everything together from scratch using the raw Messages API. It provides higher-level abstractions: a standard way to define tools and connect them to Claude, patterns for managing agent memory across turns, built-in support for human-in-the-loop interruptions where the agent pauses and waits for approval before proceeding, and multi-agent orchestration so one Claude instance can spawn and coordinate subagents.

The SDK is designed to complement the Model Context Protocol (MCP) rather than replace it. MCP defines how Claude connects to external tools and data sources; the Agent SDK defines how you structure the agent's behavior, decision loop, and lifecycle. Together they form Anthropic's recommended stack for agentic application development. Claude Code, Anthropic's agentic coding tool, is built on similar patterns, and builders can use the SDK to create domain-specific agents that apply the same architectural principles.

For practical builders, the Claude Agent SDK is the right starting point if you want to build production agents on Claude without reinventing common patterns (tool registration, error handling, loop termination conditions, agent state) from scratch. It is also the integration point for Managed Agents, Anthropic's hosted infrastructure that runs stateful agents in secure, isolated environments without you managing the compute.