Deploy the Response ChatGPT App

Use StateSet Deploy to ship the ChatGPT widget + MCP server.

Use StateSet Deploy to build and host the Response ChatGPT app and MCP server on managed Kubernetes.

1) Add a Deploy Button

Add a deploy button to your repository README:

[![Deploy to Stateset](https://deploy.stateset.com/button.svg)](https://deploy.stateset.com/deploy?repo=YOUR_ORG/YOUR_REPO)

2) (Optional) Add stateset.yaml

Create a stateset.yaml in the repo root to configure build/runtime:

name: stateset-response-chatgpt-app
description: Response ChatGPT widget + MCP server

build:
  dockerfile: Dockerfile
  context: .

runtime:
  port: 3030
  health_check: /health
  readiness_check: /ready

3) Deploy

Click the deploy button, authenticate with GitHub, and wait for the build to finish.

Your app will be available at:

https://{app-name}.{slug}.cloud.stateset.com

4) Verify the deployment

Before wiring ChatGPT to it, confirm the service is up and the MCP endpoint is answering:

APP="https://{app-name}.{slug}.cloud.stateset.com"

# The readiness probe configured above
curl -sS "$APP/ready"

# The MCP endpoint, which must return an SSE stream rather than JSON
curl -sS -N -H 'Accept: text/event-stream' "$APP/mcp" | head -c 200

A 200 from /ready and an event: line from /mcp mean the app is ready to connect. JSON or a 404 from /mcp means the MCP route is not mounted — check that the container is listening on the runtime.port declared in stateset.yaml, not on a default port.

5) Connect ChatGPT

Use the app URL as your MCP SSE endpoint in ChatGPT:

• https://{app-name}.{slug}.cloud.stateset.com/mcp

The MCP endpoint is public as soon as it deploys, and an MCP server exposes tools that act on your StateSet data. Put authentication in front of /mcp before you connect anything real to it — a deploy URL is not a secret, and ChatGPT is not the only client that can reach it.

Next steps

Response ChatGPT app guide

Building the widget and MCP server this deploys.

StateSet Deploy

The deployment service, its build pipeline and its limits.

Deploy an ACP handler

The same pipeline for a checkout handler.

MCP servers

What an MCP server exposes, and how clients connect.