Lectures

In this lecture you will build a mental “coordinate map” for ChatGPT Apps: how local dev, staging, and production differ, why the ChatGPT client is always in the cloud, how Dev Mode works in practice as a URL switch, and how to link environments with Git branches, domains, and environment variables (for example, NEXT_PUBLIC_APP_ENV and NEXT_PUBLIC_BUILD_COMMIT).

In this lecture, you will build a working “mental model” for debugging the ChatGPT App by layers: widget (React), backend/ MCP server, and the MCP protocol (JSON‑RPC). We will learn how to read Next.js and tooling logs, call tools/list/ tools/call via the MCP Inspector, and verify Dev Mode configuration (URL, /mcp path, tunnel) so you can stop debugging via a “GPT guessing game” and switch to a proper engineering process.

We break down why “random” tunnel addresses break ChatGPT Apps development, and how to switch to a stable dev URL: a named Cloudflare Tunnel with your domain or ngrok with a static domain. Plus — how to bind the domain to code via NEXT_PUBLIC_APP_URL / baseUrl, which matters for /mcp and /.well-known/openai-app, and how Dev Mode “picks up” updates without constantly reconnecting.

We break down how to deploy a ChatGPT App to Vercel without a “manual server”: connect a Git repository, configure environment‑specific env variables (Development/Preview/Production), get preview deployments for every PR, ship to production from the main branch, and perform quick rollbacks. We also discuss nuances of Next.js 16 and the MCP endpoint (serverless model, timeouts, maxDuration) and how to bind the ChatGPT Dev/Prod App to a stable HTTPS URL after deployment.

In this lecture we break down where a ChatGPT App deployment can fail (build, runtime, and configuration drift), how to read Vercel logs instead of guessing, how to protect against crashes due to process . env, why you need a redeploy after editing env, how to stabilize the MCP route /mcp, and how to roll back quickly (rollback) without making the incident worse.