Vercel adds one-click deploys for ChatGPT apps built with Next.js
ChatGPT apps can now run natively inside the OpenAI sandbox instead of a nested iframe, using Next.js server rendering and a ready-made template.
Vercel announced today that developers can build and deploy ChatGPT apps directly on Vercel using Next.js, OpenAI’s Apps SDK, and the mcp-handler library. A ready-made Next.js template deploys the whole thing in one step.
The detail that matters here is where the app actually runs: these apps work over the Model Context Protocol and now render natively inside the OpenAI sandbox rather than in a nested iframe, so a Next.js app gets to keep server-side rendering and React Server Components instead of being boxed into static markup. mcp-handler (the renamed @vercel/mcp-adapter) serves MCP’s 2026-07-28 spec while still falling back to the older Streamable HTTP transport for clients still on the 2025-era protocol. Deploys get Vercel’s usual preview-deployment and instant-rollback flow on top.
Why it matters: if you’re building a ChatGPT app, you no longer have to hand-roll hosting for an MCP server or accept an iframe’s limits on what the UI can do. You can ship it through a deployment pipeline most Next.js teams already use, with real SSR rather than a static widget.
The caveat: the template and this announcement are Next.js-specific; mcp-handler itself also supports Nuxt, SvelteKit and Hono, but native sandbox rendering is what’s being promoted here, not a general multi-framework guarantee.