Build guides

The rest of the framework is theory - what MCP is, how the agent loop works, how permissions should be designed. This section is theory's opposite. It's practice. Pick a build guide, open a terminal, and by the end you'll have a real working agent running on your machine. Three guides, three distinct agent shapes, each one teaching different parts of the stack.

The three guides, and what each one teaches.

Start with First agent if you've never built one. The other two assume you've been through that one.

The three guides.

• Your first autonomous agent in 60 minutesInstall Claude Code, wire one MCP, write a skill, run it headless. Complete beginner path, every step explained.→
• Voice agentBuild an agent that answers phone calls. Speech-to-text in, Claude thinks, text-to-speech out. Surprisingly close to production-ready.→
• Research agentSearch, extract, synthesize, report. Multi-step reasoning over the live web. This is the classic ReAct agent, done right.→

Prerequisites, the same for every guide.

• A Mac or Linux machine (Windows via WSL works fine too)
• Node.js 18+ or Bun installed
• An Anthropic API key (you'll get one by signing up; free trial credits are enough for all three guides)
• A text editor you like (VS Code, Zed, vim, whatever)

The shape every guide follows.

Open a guide, follow the steps, run the commands. By the end you'll have a real agent doing real work. Not a demo. The code you write in these guides is the same code you'd ship.