anthropics/claude-quickstarts

What it solves

This repository provides a collection of reference implementations and starter templates to help developers build applications using the Claude API. It reduces the friction of starting a new AI project by providing concrete examples of complex patterns like browser automation, computer control, and autonomous coding agents.

How it works

The repository is organized as a set of independent projects (quickstarts). Each project demonstrates a specific capability of the Claude API, such as:

• Computer Use: Tools that allow Claude to control a desktop computer or a macOS desktop with best practices for image sizing and prompt caching.
• Browser Automation: A Playwright-backed tool that enables Claude to navigate websites and interact with DOM elements.
• Managed Agents: Implementations showing how to integrate persistent sessions, generative UI components, and knowledge wikis created from document corpora.
• Autonomous Agents: A two-agent pattern (initializer and coding agent) that uses git for persistence across sessions.

Who it’s for

Developers who are using the Claude API and want to see real-world examples of how to actually implement agentic workflows, tool use, and specialized AI assistants (like financial analysts or customer support agents).

Highlights

• Computer Control: Reference implementations for both containerized and native macOS environments.
• Browser Automation: A complete implementation for navigating and interacting with websites via Playwright.
• Autonomous Coding: A multi-session coding agent that persists progress via git.
• Managed Agents: Examples of integrating with Vercel's Chat SDK and CopilotKit for generative UI.
• Knowledge Wiki: A system for distilling document corpora into a versioned memory-store for cost-effective retrieval.