Set up ChatGPT to ship code through MCP

A developer linked ChatGPT to a custom PaaS so it can launch deployments and hand back a URL without manual Docker or YAML steps.

Done with ChatGPT· Advanced· Nov 2026

What they did
Harun added MCP support to HYNA-Stage v3.0.0, his self-made PaaS. He exposed it through a custom connector for ChatGPT or Claude Desktop, with a custom OAuth 2.1 + PKCE sign-in because OpenAI’s public directory needs a payment method. When ChatGPT is asked to deploy an app, the server spins up an E2B MicroVM in isolation and sends back a live URL. The backend sits on Cloudflare and Supabase, and Harun built it from Termux on his phone.
How it went
He reports a warm sandbox in under 3 seconds, cold starts at 7 to 15 seconds, and hot reload around 300ms. He also claims 99.9% uptime over 7 days and more than 99% deployment success, without describing how he tested those numbers. The free tier is limited to 3 concurrent sandboxes, and the system relies on free tiers from Supabase, Cloudflare, E2B, and Netlify.
Tips
If you try this pattern, add a clear confirmation gate before any deploy that could affect a real environment. Free-tier limits can become the first bottleneck, so plan for sandbox count and hosted services early.

Try it yourself with ChatGPT

I want to connect [ChatGPT or another assistant] to my platform so it can deploy [project/repo] to [hosting target]. Build an MCP server that exposes a deploy tool, sets up the needed auth flow, and lets the assistant prepare and launch the release without me writing Dockerfiles or YAML by hand. Before anything is deployed, list the config, the commands, and the exact action that would be taken, then wait for my approval before you run a build, change a live service, or submit the deployment.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on dev.to ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.