Claude Code's brain is now an SDK—fine print included

What it does

The Claude Agent SDK exposes Claude Code’s capabilities as a Node.js library for building autonomous agents. Your agent can read a codebase, edit files, run shell commands, and chain those actions into complex workflows. The in-repo README is little more than a signpost, so the actual API shape and mechanics are left to Anthropic’s external documentation.

The interesting bit

This is a rebrand of the former Claude Code SDK, signaling Anthropic wants third-party developers to embed Claude’s agentic behavior into their own products. The unusual part is the legal packaging: the same Commercial Terms of Service bind you even when you ship products built on top of the SDK to your own customers.

Key highlights

• Programmatic access to Claude Code-style agent capabilities: codebase understanding, file edits, and command execution
• Ships as an npm package targeting Node.js 18+
• Replaces the previous Claude Code SDK; breaking changes are documented in a migration guide
• Collects usage feedback including code acceptance/rejections, conversation data, and /bug submissions
• Governed by Anthropic’s Commercial Terms of Service, which extends to end-user products you build on top

Caveats

• The in-repo README is a stub; most technical detail lives in external documentation
• Data collection is broad and includes conversation content, though Anthropic states it isn’t used for model training
• The Commercial Terms apply to your own customers and end users, not just your internal usage

Verdict

Worth evaluating if you need to script autonomous coding agents inside Node.js. Look elsewhere if a standard open-source license and minimal data collection are non-negotiable, because Anthropic’s Commercial Terms and built-in feedback collection apply by default.