Skip to content

anthropic-skills-tutorial

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

anthropic-skills-tutorial

README.md

layout default
title Anthropic Quickstarts Tutorial
nav_order 91
has_children true
format_version v2
source_repo https://github.com/anthropics/anthropic-quickstarts
categories
ai-agents
computer-use
tool-use
multi-turn-conversations
related_tutorials
../anthropic-code-tutorial/
../mcp-python-sdk-tutorial/
../claude-code-tutorial/

Anthropic Quickstarts Tutorial

A deep-dive into every project in the official anthropics/anthropic-quickstarts repository — computer use, autonomous coding, customer support, financial analysis, and the agents reference implementation.

GitHub License: MIT

What This Tutorial Covers

The anthropics/anthropic-quickstarts repository is the canonical starting point for building production-quality Claude-powered applications. It is not a skills/plugin system — it is a collection of five standalone quickstart projects that demonstrate the full range of Claude's capabilities:

Project What It Demonstrates
computer-use-demo/ Claude controlling a real desktop via screenshot + xdotool actions
agents/ A minimal reference agent loop with tool use and MCP integration
autonomous-coding/ Two-agent pattern: initializer + coding agent across many sessions
customer-support-agent/ Next.js chat app with Claude + Amazon Bedrock RAG knowledge base
financial-data-analyst/ Next.js app with file upload, Claude analysis, and Recharts visualizations
browser-use-demo/ DOM-aware browser automation via Playwright instead of pixel coordinates

Current Snapshot (auto-updated)

Why This Repository Matters

Before these quickstarts existed, the standard approach was to cobble together ad-hoc integrations from API documentation snippets. The quickstarts provide:

  • Working Docker environments so you can run computer use in minutes, not days
  • Reference sampling loops demonstrating multi-turn conversation management, prompt caching, and image window management
  • Concrete tool implementations showing exactly how bash, computer, and str_replace_based_edit_tool are structured
  • Production patterns like retry logic, provider abstraction (Anthropic / Bedrock / Vertex), and structured output validation

Architecture Overview

graph TD
    subgraph quickstarts["anthropic-quickstarts"]
        CU["computer-use-demo<br/>Python + Docker + Streamlit"]
        AG["agents/<br/>Python reference impl &lt;300 lines"]
        AC["autonomous-coding/<br/>Claude Code CLI + Python"]
        CS["customer-support-agent/<br/>Next.js + Bedrock RAG"]
        FA["financial-data-analyst/<br/>Next.js + Recharts"]
        BD["browser-use-demo/<br/>Python + Playwright + Docker"]
    end

    API["Anthropic API<br/>(claude-opus-4 / sonnet-4 / haiku-4)"]
    MCP["MCP Servers<br/>(optional)"]

    CU --> API
    AG --> API
    AG --> MCP
    AC --> API
    CS --> API
    FA --> API
    BD --> API
Loading

Chapter Guide

Chapter Topic Core Question Answered
1. Getting Started Setup & mental model What does each quickstart actually do and how do I run it?
2. Quickstart Architecture Project anatomy How are the five projects structured and what patterns do they share?
3. Computer Use Deep-Dive Computer use agent How does Claude control a desktop: tools, loop, coordinate scaling?
4. Tool Use Patterns Tool design How are BashTool, ComputerTool, EditTool, and custom tools built?
5. Multi-Turn Conversation Patterns Sampling loop How does the agentic loop work, and how do you manage context?
6. MCP Integration MCP How does the agents quickstart connect to MCP servers?
7. Production Hardening Reliability Prompt caching, image truncation, provider abstraction, security
8. End-to-End Walkthroughs Case studies Full traces of the customer support and financial analyst quickstarts

Prerequisites

  • Python 3.11+ and Node.js 18+ for local development
  • Docker Desktop for computer-use and browser-use demos
  • An ANTHROPIC_API_KEY from console.anthropic.com
  • Basic familiarity with async Python or TypeScript/React

Related Tutorials

Prerequisites:

Complementary:

Next Steps:

Ready to begin? Start with Chapter 1: Getting Started.

Built from the official anthropics/anthropic-quickstarts repository. All code examples are taken directly from that source.

Navigation