Master Claude Code
From Zero to Deployed

Install VS Code and the Claude Code Extension

Claude Code runs inside an IDE (Integrated Development Environment). VS Code is the recommended starting point. It is free, widely used, and the Claude Code extension installs in seconds.

• Go to Google and search VS Code, then download and install it for free.
• Open VS Code, click the Extensions icon in the left sidebar, and search Claude Code.
• Click Install on the Claude Code extension that appears.
• Alternatively, use Cursor or Windsurf as your IDE. Claude Code also runs as a standalone app.

Sign In and Open Your Project Folder

Claude Code requires a project folder to work in. Every automation lives inside its own folder, which keeps things organized and makes deployments clean.

• Click the Anthropic logo icon in the top-right corner of VS Code to open Claude Code.
• Sign in with your Anthropic account when prompted.
• In the Explorer panel (left sidebar), click Open Folder and create or select a folder for your project. Example name: youtube-analysis
• Enable Bypass Permissions mode: go to Settings, search for Claude Code, and toggle the option on. This lets the agent run without asking for approval at every single step.

Create Your claude.md System Prompt

Before building anything, give Claude Code a system prompt. This file, called claude.md, is the brain of your project. It tells the agent what it is, what tools it has, and how to behave. Without it, you get a generic agent. With it, you get a specialist.

Use Plan Mode to Describe Your Automation

Before building, switch Claude Code to Plan Mode. In this mode, the agent thinks harder, browses the web for the best tools, and asks you questions you probably haven't considered. The quality of your output depends heavily on the quality of this step.

• Click Plan Mode in the Claude Code toolbar.
• Write a clear brain dump: describe the goal, desired inputs, the format of the output, and how you want it delivered.
• Include questions like: "Should I use an MCP server or an API? Are there skills that could help here?"
• Answer any follow-up questions Claude asks (frequency, delivery email, data storage preferences, etc.).
• Review the plan it returns before approving. Read every section.

Approve the Plan and Let It Build

Once the plan is confirmed, Claude Code switches out of Plan Mode and starts building. It creates a to-do list, works through each item one by one, and writes the workflow files and Python tools automatically. Your job during this phase is to watch and supply any credentials it asks for.

• Approve the plan. With Bypass Permissions on, the agent will run without pausing at every step.
• Monitor the to-do list on screen. You can work in another window while it runs.
• When prompted, supply your API keys. Example: paste your YouTube API key directly into the chat.
• API keys should only ever be stored in .env files, never inside workflow or tool files. The agent will do this automatically if prompted correctly.

For the YouTube Analysis example, the agent built these seven tools automatically:

Add MCP Servers and Skills to Extend What the Agent Can Do

Out of the box, Claude Code is powerful. But you can make it significantly more capable by connecting MCP servers and installing skills.

• MCP Servers are like app stores for your agent. Instead of building separate API connections to Gmail, Calendar, Sheets, etc., you connect once to the MCP server and the agent handles all the endpoints automatically.
• Skills are custom instructions Claude loads dynamically based on relevance. They are not always loaded, only when the task calls for them. This saves tokens and improves consistency.
• Find public skills at the Claude Code templates library (link available in the description of the original video).
• Install a skill by copying its install command into Claude Code's chat. The agent will run the terminal command and confirm installation.
• If you repeat the same instruction to Claude Code across multiple sessions, that is a signal: turn it into a skill or add it to your claude.md file.

Run the Workflow, Review the Output, and Iterate

After building, do a full test run before deploying. Claude Code will self-heal on errors, but your feedback is what shapes the final output. This is the most important phase before going live.

• Type: "Run the [workflow name] workflow." Claude will find the right file and execute all tools in order.
• Review every output: the email, the attached PDF or PowerPoint, the Google Sheet, any charts generated.
• If something is wrong, describe it in plain language. Example: "The PDF only had two slides. It was missing the charts and the recommendations section." The agent will investigate, fix the tool, and regenerate.
• The agent updates workflow files when it finds and fixes errors, so the same mistake does not repeat.

Push to Modal and Set Your Schedule Trigger

When your workflow is tested and working, push it to Modal, which is cloud infrastructure that only charges you when automations actually run. Tell the agent what you want and it handles the entire deployment process.

• Create a free Modal account at modal.com. You get $5 free on signup and $30 more when you add a payment method.
• In Claude Code, type: "Push the [workflow name] to Modal so it runs every Monday at 6am." Supply the Modal install command from your dashboard if prompted.
• The agent creates the deployment file, sets the schedule trigger (cron), and stores your secrets as Modal secrets so nothing is exposed.
• To trigger via webhook instead of a schedule, tell the agent to deploy with a webhook trigger. Then test it using Postman to confirm it fires correctly.
• After deployment, the workflow runs in the cloud without the agent. If you want to improve it later, edit it in Claude Code and push the updated version back to Modal.