How to Run Your ChatGPT App Locally (April 2026)

You don’t need a paid ChatGPT account to build a ChatGPT App. With sunpeak, you can develop and test ChatGPT Apps entirely on your local machine using a built-in replica of the ChatGPT runtime called the Inspector. This guide covers everything from first install to connecting with the real ChatGPT.

TL;DR: Run npx sunpeak new sunpeak-app, then cd sunpeak-app && pnpm dev. Open localhost:3000 to test your ChatGPT App in the Inspector. No account, no credits, no manual refreshing.

What are ChatGPT Apps?

ChatGPT Apps are interactive UIs that run inside ChatGPT conversations. When a user asks ChatGPT to do something, your App can render a rich interface directly in the chat: forms, dashboards, carousels, data visualizations, multi-step workflows, anything you can build as a web app.

ChatGPT Apps are built on the Model Context Protocol (MCP). Your App is structured around two MCP primitives:

• Resources are HTML pages rendered in sandboxed iframes inside the chat. These are the UI your user sees.
• Tools are actions the AI model can invoke, like fetching data or running a calculation. When a tool has an associated resource, ChatGPT renders that resource as an interactive App view.

Because ChatGPT Apps use MCP, the same App also works in Claude (as a connector) and other MCP-compatible hosts. The MCP concepts explained post covers the architecture in more detail.

Setting up your local environment

You need Node.js 20 or higher. Then scaffold a project:

npx sunpeak new sunpeak-app

This creates a project with a working ChatGPT App out of the box:

sunpeak-app/
├── src/
│   ├── resources/     # UI components (React views rendered inside ChatGPT)
│   ├── tools/         # Tool handlers (actions the model invokes)
│   └── server.ts      # MCP server configuration
├── tests/
│   └── simulations/   # Deterministic test fixtures
└── package.json

Start the development server:

cd sunpeak-app && pnpm dev

This starts two things: your MCP server on port 8000 (handles tool calls and serves resources) and the sunpeak Inspector on port 3000 (a local replica of the ChatGPT App runtime). Open localhost:3000 and you’ll see your App running immediately.

Developing with the Inspector

The sunpeak Inspector replicates how ChatGPT renders MCP Apps, so you can build and test without deploying anything. It shows you exactly how your App will look and behave, including:

• All three display modes: inline, fullscreen, and picture-in-picture
• Light and dark themes
• Tool invocations with configurable mock responses
• Simulation files that load specific App states for repeatable testing

Every code change hot-reloads instantly. No manual browser refresh, no reconnecting to ChatGPT, no 4-click refresh cycle.

The Inspector also supports Claude, so you can switch hosts from the sidebar dropdown and see how your App renders in both environments from one codebase. This works because sunpeak Apps are MCP Apps that run across hosts without code changes.

Testing your App locally

Beyond manual testing in the Inspector, sunpeak includes a testing framework for automated testing. Tests run against the local Inspector using Playwright, which means they replicate real ChatGPT behavior without a paid account or API credits.

pnpm test           # run unit + e2e tests
pnpm test:visual    # visual regression testing
pnpm test:eval      # multi-model LLM evals

Simulation files in tests/simulations/ provide deterministic mock data, so your tests produce the same result every time. You can add these tests to a CI/CD pipeline so every pull request gets validated automatically. The complete guide to testing ChatGPT Apps covers unit tests, E2E tests, visual regression, and evals in depth.

Connecting to the real ChatGPT

When you’re ready to see your App in ChatGPT, expose your local MCP server with a tunneling tool like ngrok (a free account works):

ngrok http 8000

Copy the forwarding URL, then add it in ChatGPT:

1. Go to Settings > Apps > Advanced settings
2. Enable Developer mode (requires a Plus, Team, or Enterprise account)
3. Click Create app and enter your ngrok URL with the /mcp path, for example: https://abc123.ngrok-free.dev/mcp

Start a new conversation and invoke one of your tools. ChatGPT will call your local MCP server and render your App UI in the chat.

If you don’t have a paid ChatGPT account, the Inspector fully covers local development and testing. You can build and ship your App without connecting to ChatGPT until you’re ready to deploy. See the ChatGPT App without a paid account post for more on this workflow.

Deploying to production

When your App is ready, build and start the production server:

pnpm build && pnpm start

pnpm build compiles your resources into optimized bundles and your tools into production-ready modules. pnpm start launches the MCP server on port 8000 (configurable via the PORT environment variable). Deploy to any Node.js host, or use createMcpHandler for custom Express setups and createHandler for serverless platforms. The deployment guide covers production configuration in detail.

Get started with the quickstart guide, explore the ChatGPT App framework, or try the full tutorial for a deeper walkthrough.