Claude Launchpad

Claude Launchpad

**Four ways to build with Claude, and how to tell them apart.** Chat, Claude Desktop, Claude Code, and the Agent SDK each solve a different problem. This repo explains what each one is, why it exists, and when to reach for it — in plain language, with real Python you can run.

The one-paragraph version

Getting a model to answer a question is easy now. The hard part is choosing the *right* tool for what you're actually doing. Reach for the raw **API** when you want Claude inside software you're building. Open **Claude Desktop** when you want to get work done without writing glue code. Run **Claude Code** when you want real coding done in a real project. Build with the **Agent SDK** when you want to ship an autonomous agent of your own. Pick wrong and everything feels harder than it should. Pick right and it feels like the tool was made for the job — because it was.

Start here

New to all of this? Read the table below, look at the architecture diagram, then run [`examples/1_api_hello.py`](./examples/1_api_hello.py). Each example is a single, short, commented file. Full setup is at the [bottom](#setup).

The four ways in

Everything is built on the model API. The other three are progressively more capable ways of putting that model to work — from a raw call, to an app you use, to an agent that acts, to an agent you build.

	What it is	Reach for it when…	You interface via
Claude API	Direct programmatic access to the models (the Messages API)	You're putting Claude inside your own app — a bot, a summarizer, a classifier, a backend	Python / HTTP
Claude Desktop	The app: chat, Cowork, connectors (MCP), scheduled tasks	You want to get work done — f