Claude Agent SDK Starter

A minimal, batteries-included starter for building AI-powered desktop apps with the official Claude Agent SDK.

This uses @anthropic-ai/claude-agent-sdk - Anthropic's official SDK. Not a workaround. Not spoofing anything. The right way to build with Claude.

Why I Built This

I love building developer tools. Check out my other repos - I'm genuinely excited about what AI can do for developers and I want to help others build with it.

There's a lot of talk about the Claude Agent SDK but few real examples of how to build with it. This starter kit exists to:

1. Inspire builders - Show what's possible with the SDK, not just document it
2. Separate pipelines from UI - Pipelines run headlessly; the UI is just a window into them
3. Demonstrate built-in power - The SDK gives apps capabilities that would take months to build from scratch
4. Show deterministic patterns - Constrain agent calls with specific tools and skills for consistent, production-ready results

Demo: AI News Tweet Pipeline

This demo shows multi-agent orchestration - three specialized agents working together. Each agent has a single job. The output of one becomes the input to the next. This is how you build reliable AI workflows.

npm run test:ai-news-tweet

Quick Start

Prerequisites:

• Node.js v18+
• Claude Code CLI installed and authenticated

npm install
npm run dev

That's it. No API key needed - uses your Claude Code CLI authentication. Runtime binaries (bun, uv, git, msys2) download automatically on first run.

Project Structure

.claude/
  agents/ai-news-tweet/     # Agent definitions (researcher, analysis, writer)
  skills/                   # Skill definitions with scripts
  rules/                    # Behavior rules auto-loaded by Claude Code
scripts/
  downloadRuntimeBinaries.js  # Downloads bun, uv, git, msys2
  beforeBuild.js              # Pre-build hook (downloads + copies deps)
  dev.js                      # Dev mode launcher
  test-ai-news-tweet.ts       # Pipeline test script
src/
  main/                     # Electron main process
  renderer/                 # React frontend
  preload/                  # IPC bridge
  shared/apps/              # App manifests and registry
resources/                  # Runtime binaries (gitignored)

Building Your Own App

1. Copy the _template app as your starting point
2. Configure the manifest with your app's name, skills, and system prompt
3. Register your app in the registry
4. Run npm run dev

See docs/BUILDING_APPS.md for the full guide with code examples.

Authentication

The SDK routes through Claude Code CLI, using your existing Claude Pro/Max subscription. No API key required for personal development.

What's Allowed

Use Case	Allowed?
Personal development / learning	✅ Yes
Internal team tools (each user has their own Claude subscription)	✅ Yes
Distributed app where users bring their own API key	✅ Yes
Distributed app using YOUR subscription for all users	❌ No

Policy Note: Anthropic does not allow third-party developers to offer claude.ai login or share rate limits for their products. Each user must authenticate with their own credentials.

See the official Agent SDK documentation for full details.

If you distribute an app to others, each user must provide their own API key via ANTHROPIC_API_KEY.

Alternative Providers

The SDK officially supports these Claude providers:

Provider	Environment Variable
Amazon Bedrock	CLAUDE_CODE_USE_BEDROCK=1 + AWS credentials
Google Vertex AI	CLAUDE_CODE_USE_VERTEX=1 + Google Cloud credentials
Microsoft Foundry	CLAUDE_CODE_USE_FOUNDRY=1 + Azure credentials

Custom Endpoints

The SDK supports any Anthropic-compatible API via ANTHROPIC_BASE_URL. Z.AI GLM is included as an example:

Note: Z.AI uses Zhipu AI's GLM models, not Claude. This is useful for cost savings or when Claude is unavailable in your region, but you're running a different model.

Configure in Settings or via environment:

GLM_API_KEY=your-zai-api-key

Other Anthropic-compatible endpoints work the same way using ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN.

Runtime Binaries

The Claude Agent SDK needs certain tools to run scripts:

• bun - JavaScript runtime for skill scripts
• uv - Python package manager for Python skills
• git + msys2 (Windows) - Shell utilities (bash, grep, sed)

These download automatically on first npm run dev. No manual installation required.

Build for Distribution

npm run build:win    # Windows installer
npm run build:mac    # macOS app

Output goes to dist/ with all runtime binaries included.

Commands

# Development
npm run dev              # Start dev mode with hot-reload
npm run build            # Build for production
npm run typecheck        # Type check
npm run lint             # Lint code

# Testing
npm run test:ai-news-tweet   # Run pipeline test
npm run test:bypass-auth     # Test SDK authentication
npm run test                 # Run unit tests

# Build for distribution
npm run build:win        # Windows installer
npm run build:mac        # macOS app

Troubleshooting

See docs/TROUBLESHOOTING.md for common issues and solutions.

Questions & Compliance

If you have questions about SDK usage, authentication, or compliance:

1. Open an issue on this repo - I'm happy to help
2. Check the official docs - Agent SDK Overview
3. Contact Anthropic - For branding or commercial licensing questions, reach out to Anthropic's sales team

This starter kit follows Anthropic's published guidelines. If you believe something here doesn't comply with their terms, please open an issue and I'll address it.

License

MIT

---

Built with the Claude Agent SDK | Official Documentation