The WorkshopArchived

Build something.

The Workshop is three practical guides that end in a working setup: standing up an AI agent, installing a skill, and connecting an MCP server. Each guide is written to be followed once, start to finish, rather than skimmed. The volume is archived as written and kept at its original URL, with the archive date stated on the page.

Archived 2026-08-11

This volume is kept for reference and is no longer maintained. Commands and package names were correct when it was written and have not been rechecked since, so run each one against the current documentation before you trust it. It stays at its original URL. The three volumes are introduced together on the learn hub.

Contents

Three guides. Real commands. Working results.

Start at the top and work down, or jump to what you need. Each guide is written to be followed once, start to finish, rather than skimmed.

01Set up your AI agent
02Install and use skills
03Connect MCP servers

01Guide

Set up your AI agent.

We’ll set up Claude Code — Anthropic’s agentic coding tool. It runs in your terminal, reads your entire codebase, and writes real code. You need Node.js 18+ and an Anthropic API key.

Install Node.js if you do not have it
```
brew install node
```
On macOS with Homebrew. For other systems, visit nodejs.org.

Install Claude Code globally

npm install -g @anthropic-ai/claude-code

Set your API key
```
export ANTHROPIC_API_KEY=sk-ant-your-key-here
```
Get your key from console.anthropic.com. Add this to your shell profile (~/.zshrc or ~/.bashrc) to persist it.
Run it in a project
```
cd your-project && claude
```
That’s it. The agent starts, reads your project, and waits for instructions.
Give it a task
```
"Read the README and summarize what this project does"
```
Start simple. As you build trust, give it bigger tasks: “Add a dark mode toggle”, “Write tests for the auth module”, “Refactor this component to use hooks”.

The agent learns your codebase as it works. It reads files, checks types, runs tests. It’s not generating code in a vacuum — it’s working inside your project.

02Guide

Install and use skills.

Skills give your agent specialized capabilities. A skill for deployment knows the exact steps. A skill for testing knows the framework conventions. Skills are loaded on demand — they don’t bloat your agent when unused.

Walkthrough · install your first skillAll recordings

Understand where skills live

~/.claude/skills/    global skills, available in every project
.claude/skills/      project-local skills, scoped to this repo

Create a skill file

# ~/.claude/skills/deploy.md
---
name: deploy-to-vercel
description: Deploy the current project to Vercel
---
When asked to deploy, run `npx vercel --prod`.
Confirm the deployment URL with the user.
If it fails, check for build errors first.

Use the skill
```
/deploy
```
Type the slash command in Claude Code. The agent loads the skill and follows its instructions. You can also just say “deploy this project” and the skill activates by matching the trigger.
Install community skills
Community skills are markdown files you download and drop into your skills directory. Browse curated collections, copy the file, and the capability is available immediately.

Good skills are specific and opinionated. “Deploy to Vercel” is better than “deploy anywhere”. The specificity is what makes the agent reliable.

03Guide

Connect MCP servers.

MCP servers give your agent access to external tools — databases, APIs, browsers, file systems. Once connected, the agent can use these tools as naturally as reading a file.

Open your Claude Code settings
```
claude config
```
Or edit ~/.claude/settings.json directly.

Add an MCP server (example: GitHub)

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "@modelcontextprotocol/server-github"
      ],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token"
      }
    }
  }
}

Restart Claude Code
Exit and re-launch Claude Code. It discovers MCP servers on startup. You’ll see the server’s tools listed in the available tools.
Use the tools
```
"List my open pull requests on this repo"
```
The agent calls the GitHub MCP server’s tools automatically. No special syntax needed — just describe what you want.
Add more servers
- filesystem
  Read and write files outside the project
- postgres
  Query and manage databases
- supabase
  Full Supabase platform access
- browserbase
  Cloud browser automation

Each MCP server you add expands what your agent can do. Start with one or two and add more as you need them. The configuration is the same pattern every time.

Source: content/learn/index.ts — the same entry the learn hub reads
Method: Archive date read from the volume entry; step numbers come from a CSS counter, never typed
Built: 2026-08-27 22:54 MT