What is Agent SDK?

Anthropic's docs describe an agent as software that plans its own steps and uses tools to work with files and commands. The Agent SDK provides the same tools, agent loop and context management that run Claude Code, as a library for Python and TypeScript. The docs say it runs the Claude Code binary, which makes it a way to embed that agent in an application the developer operates.

The overview lists the capabilities carried over from Claude Code: built-in tools for files, commands and web search, hooks that run custom code at points in the agent lifecycle, subagents for focused subtasks, connections to MCP servers, permission controls, sessions that can be resumed or forked, and skills, commands and memory loaded from the project's .claude directory. Plugins can bundle several of these.

The docs contrast it with the Claude Code CLI for interactive terminal work, the Client SDK where the developer writes the tool loop, and Managed Agents where Anthropic hosts the agent. For other languages, the CLI can be run as a subprocess in print mode with JSON output. One stated limit: third-party developers may not offer claude.ai login or its rate limits in their products without prior approval, so API key authentication is used instead.