Most SDK example repos stop at a "hello world" and a link to the docs. This one ships eight self-contained apps that each answer a harder question: what does a real agent built on the Claude Agent SDK actually look like in code? The value here isn't the SDK itself — that's documented elsewhere — it's watching the same primitives (subagents, tool use, sessions, streaming) composed into shapes you'd recognize from production work.
What Sets It Apart
- Range over depth. The demos span an IMAP email assistant, a multi-agent research system with parallel subagents and web search, an Excel/spreadsheet agent, a React + Express chat UI with WebSocket streaming, and a .docx resume generator — so you can pattern-match to your own use case instead of extrapolating from one toy.
- Two SDK eras side by side. "Hello World" and "Hello World V2" contrast the original API with the V2 Session API (multi-turn conversation, session persistence), making the migration path concrete rather than abstract.
- Mostly TypeScript, some Python. Roughly 89% TS and 9% Python means the patterns transfer to the two stacks most teams actually build agents in.
- First-party, so the patterns track the SDK. Maintained by Anthropic alongside the SDK, the examples tend to reflect current idioms rather than drifting community conventions.
Who It's For
Great fit if you're evaluating the Claude Agent SDK and want working reference implementations to read, fork, and strip down — especially the research and email agents, which show genuine multi-step, multi-tool orchestration. Look elsewhere if you need the SDK API reference (read the docs), or a deployable template: the README is explicit that these are local-development demos, not hardened for production or scale.
