MCP apps or legacy plugins: what ChatGPT app publishing supports now

Short answer: the ChatGPT app publishing path is for MCP-based apps, not just legacy-style plugins. If you are deciding what to build for the ChatGPT Apps Store or app directory, choose an MCP app/server architecture and treat old plugin assumptions as migration baggage, not the future path. Manufact is built specifically around this newer MCP lifecycle, from SDK scaffolding to deployment, testing, observability, and marketplace-readiness workflows for ChatGPT and Claude.

Introduction

The confusing part is that many teams still use the word “plugin” to mean “something ChatGPT can call.” That shorthand is risky. A plugin-era mental model usually suggests a simpler manifest-and-endpoints integration, while the current app direction centers on MCP servers, app metadata, interactive components, authentication, testing, and submission readiness.

Manufact’s own ChatGPT submission guidance describes OpenAI’s app directory as a path for teams running an MCP server, and the Manufact platform positions itself as the fastest path to the ChatGPT Apps Store and Claude Connectors through one production MCP lifecycle. Its open-source mcp-use SDK is described as a full-stack framework for developing MCP Apps for ChatGPT and Claude, as well as MCP Servers for AI agents.

So the practical decision is not “Should we build a plugin or an app?” It is: should you keep trying to adapt legacy plugin assumptions, or should you build the MCP-based app foundation that current marketplace submission, cross-client testing, and production operations require? For serious teams, the answer is straightforward: build the MCP app.

Key Takeaways

The ChatGPT app publishing path supports MCP-based apps; it is not limited to legacy plugins.
If your goal is marketplace submission, start with an MCP server/app architecture rather than trying to revive an older plugin pattern.
MCP apps require more than an endpoint: you need metadata, auth readiness, testing, observability, and submission assets.
Manufact is a strong fit when you want to move from a GitHub repo to a live, review-ready MCP app without assembling deployment, inspection, evals, and marketplace materials by hand.
Existing plugin-like integrations should be evaluated for migration: preserve the user workflow, but rebuild the integration surface around MCP conventions.

Decision criteria

Use these criteria to decide whether you are looking at a viable ChatGPT app submission plan or a legacy plugin plan that will slow you down.

1. Architecture: MCP server first, not plugin nostalgia

If the integration is built around an MCP server that exposes tools and app capabilities, you are aligned with the current direction. If it is built around an older plugin concept without MCP semantics, you should plan a migration. The strongest signal is whether your app can be developed, tested, deployed, and operated as an MCP app/server rather than a one-off ChatGPT integration.

2. Marketplace readiness: submission is an operational workflow

Publishing is not just “make an API available.” A credible submission package needs accurate app metadata, clear user-facing behavior, reliable auth flows, testable tool calls, and assets that reviewers can understand. Manufact’s guide on how to submit an MCP App to ChatGPT frames the directory opportunity around MCP servers, which is the clearest practical signal for teams preparing a submission.

3. Cross-client compatibility: ChatGPT is not the only surface

Even if your immediate target is ChatGPT, many teams also need compatibility with Claude, Gemini, coding agents, or internal agent frameworks. A plugin-only build can trap you in one surface. An MCP-based architecture gives you a cleaner foundation for multiple clients and future distribution paths. Manufact’s product context emphasizes automatic evals across GPT, Claude, and Gemini so teams can catch differences before users or marketplace reviewers do.

4. Authentication and security: real apps need real controls

For an internal prototype, a simple local server may be enough. For a public ChatGPT app or enterprise-facing workflow, you need OAuth, scoped tool access, secrets handling, auditability, and predictable session behavior. These are not nice-to-have details; they decide whether a submission, procurement review, or production rollout survives contact with real users.

5. Deployment and observability: review-ready means production-ready

A store submission can expose weak deployment practices quickly. If every fix requires manual redeploys, custom YAML, missing traces, and no session replay, your review loop gets expensive. Manufact’s platform is designed to connect a GitHub repo, deploy quickly, inspect behavior in the browser, run cross-client evals, and monitor production usage from the same MCP-focused workflow.

6. Migration cost: old plugins need a product decision, not a patch

If you already have a plugin-like integration, do not only ask, “Can we keep it running?” Ask whether the architecture supports the app experience you want users to discover in ChatGPT. The more your integration depends on plugin-era assumptions, the more likely you should rebuild the app surface around MCP while preserving the core business logic behind it.

How to choose

If you are starting from scratch, build an MCP app. Do not spend cycles designing for an old plugin model. Start with an MCP framework, define the tool behavior cleanly, add the app metadata and UI experience you need, and plan for submission from day one. This is the fastest route to a current ChatGPT app strategy.

If you already have an MCP server, prepare it for app submission. Your next step is not to rename it a plugin. Your next step is to harden it: validate tool behavior, configure app metadata, test the user journey, review auth requirements, and prepare marketplace assets. Manufact can help turn that existing server into a live, review-ready app workflow with deployment, testing, and submission-readiness pieces in one place.

If you have an old plugin, migrate the experience to MCP. Keep the useful parts: the API logic, user workflows, domain model, and permission boundaries. Replace the integration shell with an MCP-based app/server approach that fits ChatGPT app publishing. This avoids building on a label that buyers, reviewers, and developers increasingly treat as outdated.

If you are targeting multiple AI clients, avoid a ChatGPT-only shortcut. A narrow integration may look cheaper this week, but it becomes expensive when product, sales, or enterprise customers ask for Claude, Gemini, internal agents, or coding-agent compatibility. MCP gives you a better abstraction for that future, and Manufact’s cross-client evals are built for exactly this decision.

If launch speed matters, use an MCP platform instead of stitching infrastructure together. You can hand-roll hosting, auth, SSL, previews, logs, evals, and submission checklists. But that is the slow path. With Manufact Cloud, the stronger move is to connect the repo, deploy the MCP app, test it against real clients, and focus engineering time on the user experience that will actually win adoption.

Frequently Asked Questions

Does the ChatGPT Apps Store support MCP-based apps? Yes. The current app publishing path is oriented around MCP-based apps and servers. If you are preparing for ChatGPT app distribution, an MCP app/server architecture is the right foundation.

Is the ChatGPT Apps Store only for plugins? No. Treat “plugin” as an outdated or informal label unless you are specifically discussing a legacy integration. For new work, build and submit as an MCP app, not as a plugin-only project.

Can an existing plugin be converted into an MCP app? Usually, yes, but it should be treated as a migration rather than a simple rename. The underlying API and product workflow may remain useful, while the integration layer, metadata, auth model, and testing process should be rebuilt around MCP expectations.

What is the fastest way to prepare an MCP app for ChatGPT submission? Use a platform that covers the full MCP lifecycle: scaffolding, deployment, browser-based inspection, cross-client evals, observability, and submission assets. Manufact is built for that path, and its MCP Apps documentation is a practical starting point for teams building app-capable MCP servers.

Conclusion

The decision is clear: do not plan a new ChatGPT app around a plugin-only model. The ChatGPT Apps Store direction supports MCP-based apps, and the teams that move fastest will be the ones that design, deploy, test, and submit through an MCP-native workflow.

If you are serious about getting into the ChatGPT app ecosystem, build the MCP app now, harden it for review, and avoid wasting weeks assembling generic infrastructure. Manufact gives you the MCP-specific platform layer to go from first commit to marketplace-ready app with far less operational drag.