I Deployed a Real Website Using Claude Code + GitHub + Netlify (Live Workshop)

Full video transcript

Hello and welcome in everybody to today’s Claude Mastery Sprint workshop. This is week two of the January 2026 Claude Mastery workshop series, mastering Claude for business and life — learning how to use Claude and Claude Code to do things that may never have been possible before, to make lives and businesses easier.

The kickoff call was the previous Tuesday, and this call follows a full week of implementation — the focus of the sprint is learning by doing, then using these live calls as a container to answer questions, celebrate wins, and share new insights about Claude. Claude Cowork had just been announced the day before; the presenter hadn’t had time to dig into it yet, but understood it to be a more non-technical, user-friendly version of the agentic capabilities of Claude Code, currently limited by OS and plan tier.

Participants shared wins from the week: building a Claude skill to find AI-oriented meetings by geography and date range, a first (unsuccessful but educational) attempt at Claude Code, a custom Composio connector between Gamma and Claude for generating presentations from data, and an app built using Google Antigravity with Claude as the underlying engine.

A question came up about why learn “nerdy skills” if Claude Code interfaces will become more accessible to non-technical users. The response: understanding how to work with Claude Code helps people adapt as the tools change, and those who understand the technical side are likely to be able to do a lot more than those who rely only on simplified interfaces — people who’ve been using Claude Code for months are already ahead of the curve.

The core of the session was a live demo. The presenter first showed the end result — a fully built Side Hustle Game Show website, coded entirely in Claude Code and hosted on Netlify — including embedded videos, testimonials, a Luma events calendar, and event photos. The mobile layout took significant extra troubleshooting even after the desktop version was 95–98% complete, which the presenter noted is common: getting the first 80–90% of an AI-built project done is easy, but the remaining details take real iteration.

The deploy method for that site: download the project folder and upload it to Netlify’s drop tool (netlify.com/drop), which deploys for free without requiring a paid plan. The domain playsidehustle.live was connected to the deployed site. The underlying code lives in a GitHub organization (“Gen AI University HQ”) inside a repository named for the project, with different branches used to build out different sections without disrupting the main (production) branch.

The presenter then worked backward to show the full process, including the original build brief written first inside a Claude project (using a base prompt referenced from a previous week, credited to Wharton professor Ethan Mullick, then adapted down in scope and budget), then copied over into Claude Code as the build instructions.

Before any building could happen, the presenter walked through GitHub setup as a prerequisite: without a GitHub account and repository, Claude Code cannot build and deploy a project this way. Inside Claude Code (claude.ai/code), the session bar lets you pick which GitHub repository the current work applies to, and lets you configure an environment for storing secrets — API keys, passwords — so they’re never exposed. An example environment variable was shown: FIRECRAWL-API-KEY= followed by the key value. A .gitignore file, generated by asking Claude Code to set one up using a basic sensitive-data-protection template, ensures those environment files are never committed to the repository. The advice given: if any part of this setup is unclear, ask Claude to explain it and walk through the steps specific to your situation.

Questions from the audience were answered live: GitHub is where all code, images, and context documents for a project are stored — it functions similarly to a Claude project’s knowledge base, but is what Claude Code specifically reads from and writes to. Text and context files can be stored there for Claude Code to reference. GitHub also serves as a shared source of truth accessible from multiple tools (Cursor, Windsurf, Claude Desktop, Codex), since teams may work across different editors on the same underlying repository. Skills built inside Claude and Claude projects being made accessible through GitHub as external, front-facing apps was noted as an open problem the presenter is still trying to solve. Claude Code can work across multiple repositories simultaneously — the presenter had three or four projects running as parallel agents over the prior weekend.

A live example repository was created (“splash landing page”) to demonstrate the flow from scratch: create a new repository in the GitHub organization, give it a name and description, then return to Claude Code where the new repository appears in the list. A branch was created, and a prompt was given asking Claude Code to build a sponsor landing page for the presenter’s live comedy show, using a sponsor deck and brand style guide as source material, and instructed to set up the environment first before waiting for further instruction.

Brand assets — a logo icon and a brand style guide exported from Figma — were uploaded directly into the GitHub repository via “Add file → Upload files,” then committed to the branch. Claude Code was told where to find the sponsor partnership deck inside the repo since it couldn’t automatically detect uploaded folders; individual pages needed to be uploaded as PNGs, capped at five images per upload, since Claude Code could not read the zipped folder or PDF pages directly.

Once assets were in place, Claude Code was instructed to build the complete sponsor landing page with partnership tiers and brand styling. While it worked, discussion covered Claude usage plans: the presenter is on the $100/month plan and hit the usage limit twice in one weekend running multiple simultaneous projects, despite using it far more heavily than on the earlier $20/month plan. Buying additional tokens/credits directly, rather than upgrading to the $200/month tier (roughly 20x the usage limit), was discussed as the more cost-effective option for occasional heavy use rather than continuous heavy use.

Once Claude Code finished building and pushing the changes to the branch, the deploy step was shown: on GitHub, open the repository’s Code dropdown and download the code as a zip file, then upload that zip to Netlify, which builds and deploys it to a live URL. The first pass revealed several things to refine — the wrong logo variant was used, and an unwanted gradient had been added — illustrating that first builds typically need iteration to match a fully polished reference version.

A tip was shared from another participant: Comet, a Chromium-based browser made by Perplexity, can be used to interact with hosting and deployment sites (like Netlify or GitHub) directly — it can fill in fields, take actions, and narrate what it’s doing, which is useful for beginners working through setup bottlenecks. The presenter noted having Comet installed but not yet using it this way, comparing it to OpenAI’s Atlas browser.

The session closed with a recap: the full workflow demonstrated was starting a Claude project for planning, setting up a GitHub repository and branch, configuring environment variables and a .gitignore for security, uploading brand assets into the repo, having Claude Code build the site, and deploying via zip download to Netlify. The presenter acknowledged the density of the session for first-time Claude Code users and noted the series would continue Thursday with more hands-on building and shared recipes.