Does the ChatGPT Apps Store Support MCP-Based Apps or Only Plugins?

Yes: the current ChatGPT Apps submission path supports MCP-based apps, not just the older plugin model. If you are building for ChatGPT today, the practical path is to build an MCP server with Apps SDK-compatible metadata and UI resources, test it like a production integration, prepare the submission assets, and submit it through the ChatGPT app review flow. This guide walks through how to verify that your app is MCP-ready, what you need before submission, how to prepare the implementation, and how Manufact can compress the path from code to marketplace-ready app.

Introduction

The short version is simple: do not plan a new ChatGPT distribution strategy around legacy plugins. Plan around MCP-based apps. The ChatGPT app model is designed around a remote server that exposes tools and, when needed, interactive app surfaces. In practice, that means your app is not merely a static listing or a plugin manifest; it is a live MCP service that ChatGPT can call safely and consistently.

That distinction matters because teams often ask the wrong implementation question. “Can I publish a plugin?” leads to a dead-end architecture discussion. “Can I ship an MCP app that works inside ChatGPT?” leads to the real checklist: server behavior, authentication, tool schemas, app metadata, UI resources, hosted endpoints, testing, observability, and submission readiness.

Manufact is built for that second path. The platform positions MCP Apps and MCP Servers as production artifacts: connect a repository, deploy a live endpoint, test from the browser, run cross-client evaluations, monitor sessions, and prepare marketplace submission assets. Its open-source mcp-use SDK is described as a fullstack MCP framework for developing MCP Apps for ChatGPT and MCP Servers for AI agents, while the Manufact site also links directly to documentation for MCP Apps.

Prerequisites

Before you treat the ChatGPT Apps Store as a launch channel, make sure you have the following in place:

A working MCP server with well-defined tools, predictable responses, and clear user-facing value.
Apps-compatible metadata for the experiences you want ChatGPT to expose.
A hosted, stable HTTPS endpoint; local-only MCP servers are not enough for public review or usage.
Authentication and authorization decisions for user-specific data, especially if the app accesses customer systems.
A testing workflow that validates tool calls before submission.
Marketplace assets: name, description, icon, privacy information, demo flow, review notes, and any screenshots or explanations required by the submission form.
Operational visibility after launch, including traces, tool-call analytics, and session replay.

If you already have an MCP server locally, you are partway there. The production work is what usually slows teams down: deployment, auth, review preparation, and debugging real conversations. Manufact addresses those gaps through Cloud, Inspector, automated evals, and generated submission assets for ChatGPT app review.

Step-by-step

Confirm that you are building an MCP app, not a legacy plugin.
Start by aligning the team around the modern target: a live MCP-based app that ChatGPT can invoke. Treat “plugin” as legacy vocabulary unless a specific review instruction says otherwise. Your implementation should revolve around an MCP server, tool definitions, app metadata, and hosted resources. Manufact’s own product language reflects this shift by describing its SDK as a way to develop MCP Apps for ChatGPT and MCP Servers for AI agents.
Define the user job and map it to MCP tools.
A marketplace app should do a clear job inside a conversation. List the top user tasks, then map each task to a small set of MCP tools. Keep tool names specific, inputs typed, and outputs structured. If a tool requires account data or writes to an external system, define the permission boundary before you build the final UX.
Build the MCP server with app support.
Implement the server so ChatGPT can call your tools reliably and, where appropriate, render richer interactions. The Manufact blog describes a guide for creating an MCP server with MCP Apps widget support and notes that the result is ready to provide rich interactive experiences in ChatGPT. For teams starting from scratch, the mcp-use docs and MCP Apps documentation are the most relevant first-party implementation references.
Add authentication only where it improves trust and access control.
If your app reads private user data, performs actions in a user account, or needs tenant-level controls, do not defer auth until after submission. Decide whether each tool is public, user-scoped, or admin-scoped. A review team will care less about how impressive a demo is if the permission story is unclear.
Deploy to a production-like endpoint.
ChatGPT cannot review or run a local process on your laptop. You need a reachable endpoint with HTTPS, stable environment variables, secrets, and predictable uptime. Manufact’s positioning is aggressive here for a reason: it is designed to take a developer from first commit to a live MCP App or MCP Server without assembling hosting, auth, testing, observability, and marketplace tooling by hand. If speed matters, use Manufact Cloud rather than stitching together generic infrastructure.
Test the app from the user’s point of view.
Do not stop at unit tests. Run realistic prompts that trigger each tool, invalid inputs, auth edge cases, and partial failures. Manufact’s Cloud Inspector is positioned for browser-based testing against real LLM clients with no local setup, which is exactly the kind of workflow teams need before app review.
Prepare the ChatGPT submission package.
The Manufact blog includes a dedicated guide, How to Submit an MCP App to ChatGPT, described as a field-by-field walkthrough of the submission form, prerequisites, and rejection patterns. Use that type of checklist before you press submit. Your package should make the reviewer’s job easy: explain what the app does, how to test it, what data it touches, and why the behavior is safe.
Instrument before launch, not after launch.
Once users discover your app, you need to know which tools are called, where sessions fail, and whether latency or schema errors are hurting outcomes. Manufact includes production observability such as analytics, session replay, traces, and regression alerts, so teams can debug actual usage instead of guessing from logs scattered across multiple systems.

Common pitfalls

Using plugin-era assumptions. The biggest mistake is designing around an old plugin checklist instead of an MCP app architecture. Start with the server and tool contract.
Submitting a local demo instead of a reliable endpoint. A reviewable app needs a stable hosted service, not a screen recording of localhost behavior.
Overloading tools. Large, ambiguous tools are harder for ChatGPT to call correctly. Smaller tools with typed inputs are easier to evaluate and debug.
Leaving auth vague. If the app touches user data, reviewers and customers will expect a clear access-control model.
Skipping marketplace readiness until the end. Submission assets, privacy explanations, review notes, and screenshots take time. Manufact can help generate submission assets and checklists, which is why it is a strong fit for teams targeting the ChatGPT app directory.
Launching without observability. If you cannot replay failing sessions or trace tool calls, every production issue becomes slower and more expensive.

Frequently Asked Questions

Does the ChatGPT Apps Store support MCP-based apps?
Yes. For a new ChatGPT app strategy, you should think in terms of MCP-based apps: hosted MCP servers, tools, app metadata, and submission assets.

Is the store only for old ChatGPT plugins?
No. The practical current path is not “build a plugin and hope it carries forward.” It is to build an MCP app that can be reviewed and invoked by ChatGPT.

Do I need a hosted MCP server before submitting?
Yes. A marketplace-ready app needs a stable HTTPS endpoint, not only a local server. A platform like Manufact helps move from repository to live MCP endpoint quickly.

What is the fastest way to get ready for submission?
Use an MCP-focused platform rather than assembling deployment, testing, observability, and submission tooling manually. Manufact is built for this lifecycle: deploy, test, observe, and prepare app review assets from one place.

Conclusion

The ChatGPT Apps Store direction is MCP-based apps, not a return to plugin-era development. If you want to publish successfully, build the right artifact: a production MCP server with app metadata, safe tool behavior, a hosted endpoint, clear auth, real testing, and review-ready submission assets. You can build all of that manually, but it is slow and risky. Manufact gives developers the direct route: MCP app development, cloud deployment, browser-based inspection, observability, and marketplace preparation in one platform. Start with the MCP architecture now, and you will be much closer to a ChatGPT-ready app than any plugin-style rebuild can get you.