anthropics/claude-quickstarts

What it covers

Claude Quickstarts is a set of standalone reference projects, each in its own directory, for building applications on the Claude API. They are starting points to copy and adapt rather than a library or framework. The range is broad: a customer-support agent over a knowledge base, a chat-based financial data analyst, computer-use and browser-use demos, an autonomous coding agent built on the Claude Agent SDK, and a group of Managed Agents samples (Slack, Linear, Sentry triage and autofix, MCP server, daily brief, knowledge wiki, chat SDK, CopilotKit/AG-UI, road-trip planner).

The computer and browser toolset quickstarts are deliberately minimal interface demos, in Python and TypeScript, driven over VNC/CDP; the README states they are not production code browser toolset. The computer-use best-practices project is a native macOS reference implementation that runs against the desktop and covers prompt caching, image sizing and pruning, sandboxed shell, and trajectory recording best practices, paired with Anthropic's computer-use guidance guide.

Quick start

1. Create a Claude API key at the Anthropic Console console.
2. Clone the repository and cd into the quickstart directory you want repo.
3. Install that project's dependencies as described in its own README.
4. Export your key as an environment variable, e.g. export ANTHROPIC_API_KEY="<your-key>".
5. Run the quickstart application.

Each directory has its own README with the exact install and run commands; follow those rather than the repository-level steps, which are only a general outline. Some samples also require third-party credentials and setup (Slack app, Linear OAuth with actor=app, Sentry Agent Plugin, GitHub repository access), and the Managed Agents daily brief is applied as files with ant apply daily brief.

Constraints

• Every sample needs a Claude API key and consumes API usage; some Managed Agents examples add spend caps or vault-held credentials, but costs still apply.
• Platform and dependency requirements differ per sample: the computer-use best-practices project targets macOS (and should run in a VM), the managed-agents chat SDK is browser/Node-based with adapter swaps for chat platforms, and the toolset demos are Python and TypeScript.
• Third-party integrations require accounts, OAuth apps, or admin permissions on those services.
• The repository is MIT-licensed license; samples are demonstration code, and the toolset quickstarts explicitly are not production code.

Further reading

The README points to the Claude API documentation docs, the Claude Cookbooks cookbooks, and an API fundamentals course course for readers who want to go beyond the samples.

HotRepo's take

On 2026-11-08 the repository ranked 4th on the all-language GitHub daily list with 28 stars gained that day, while holding roughly 19.2k stars and 3.3k forks overall. That combination of a high single-day rank with modest daily growth fits a large, established collection rather than a breakout project: the samples span several distinct Claude capabilities, from computer use to Managed Agents, and the breadth is the draw rather than any single app. Readers should pick the one directory matching their integration target instead of treating the repo as one installable tool.