Anthropic's official starter pack for Claude apps
Six reference implementations showing how to move from API key to deployable agent without drowning in boilerplate.

What it does This repo bundles six self-contained quickstarts—customer support, financial analysis, computer use, browser automation, and an autonomous coding agent—each with its own setup instructions and dependencies. The pitch is simple: clone, add your API key, and have something running that you can actually extend rather than throw away.
The interesting bit The “Computer Use Best Practices” quickstart is the standout. It runs natively on macOS (the README sensibly warns: use a VM) and explicitly demonstrates cost-control patterns—prompt caching, image pruning, batched tool calls, server-side compaction—that most toy demos ignore. It pairs with an actual Anthropic blog post, suggesting someone internal cares about making computer use affordable enough to ship.
Key highlights
- Customer support agent with knowledge-base integration
- Financial data analyst with interactive visualizations via chat
- Containerized computer-use demo supporting the
computer_use_20251124tool version (zoom actions included) - Native macOS computer-use reference with trajectory recording and sandboxed shell
- Playwright-backed browser automation tool for DOM interaction and form filling
- Two-agent coding pattern (initializer + coder) with git-persisted progress across sessions
Caveats
- The README is vague on what “deployable” actually means—no mention of hosting, Docker, or cloud setup beyond the containerized computer-use demo
- Each quickstart is siloed; there’s no shared library or abstraction layer, so expect to copy-paste patterns between projects
Verdict Worth bookmarking if you’re evaluating Claude for a specific use case and want to see canonical patterns before building your own scaffolding. Skip it if you need production-grade infrastructure or multi-provider portability; this is Anthropic-flavored sample code, not a framework.