// the find
anthropics/claude-quickstarts
A collection of projects designed to help developers quickly get started with building deployable applications using the Claude API
A grab-bag of reference apps from Anthropic showing how to wire the Claude API into real app patterns — customer support bots, computer/browser-use agents, and a growing set of Managed Agents integrations (Slack, Linear, Sentry, CopilotKit, etc.). It's aimed at developers who want a working skeleton to copy from rather than a library to install.
Each quickstart is a real, runnable app (Next.js, Streamlit, or a plain Python worker) rather than a stripped-down snippet, so you get auth, streaming, and UI wiring already solved. The computer-use and browser-use demos in particular show non-trivial patterns — tool result batching, prompt caching, coordinate scaling for screenshots — that are annoying to get right from the docs alone. The newer managed-agents/* additions are a good look at session-based agent patterns (webhook bridges, vault-injected credentials, multiagent handoff) that aren't documented anywhere else as concretely.
It's a monorepo of loosely related projects with no shared versioning or dependency management, so a breaking change in one quickstart's package.json doesn't tell you anything about the others — expect to vendor just the folder you need. Quality is uneven: some projects (computer-use-best-practices) have real test suites and CI, others (financial-data-analyst) have none. The sheer number of managed-agents examples (nine and counting) makes the repo read more like a feature showcase for Anthropic's agent platform than a stable set of quickstarts you'd build production code on top of.