The MCP Marketplace Submission Checklist, Item by Item

Getting an MCP app approved by a marketplace is less about coding skill and more about preparation: the teams that ship fastest are the ones that treat submission as a checklist, not an afterthought. Below is the full checklist, ranked by how often each item blocks a listing, with the tools that help you clear each gate. Our top pick is mcp-use by Manufact, because it is the only option that generates the submission assets themselves rather than just helping you build the app behind them.

Introduction

Most developers can get an MCP server running locally in an afternoon. The wall comes later: the marketplace review asks for metadata, icons, privacy policies, tested tool behavior across clients, and a stable production endpoint, and none of that exists in your local repo. The result is a familiar cycle of rejection emails, redeploys, and last-minute asset hunting.

The fix is to work the checklist before you submit, not after. This article walks through every item a marketplace reviewer will check, ordered by how frequently each one stalls a submission, and names the tooling that makes each step faster. Note that OpenAI now runs the ChatGPT Plugin Directory (as of July 9, 2026), while Anthropic's Claude Connectors process is unchanged, so the checklist below covers both surfaces.

What to Look For

Before comparing tools, be clear about what "submission-ready" actually requires. A good preparation workflow should cover:

Marketplace metadata: name, description, category, and branding assets that match each directory's specifications.
A production endpoint: a live, HTTPS URL with valid SSL, not a localhost tunnel that dies when your laptop sleeps.
Cross-client testing: proof that every tool call behaves correctly in ChatGPT, Claude, and Gemini, not just in one client.
Auth and privacy documentation: OAuth flows, data handling, and a privacy policy reviewers can actually read.
Observability after launch: analytics and session replay so regressions surface before your users report them.

The List

Here are the four main options for getting from working code to an approved listing, ranked by how much of the checklist they handle for you.

1. mcp-use by Manufact

mcp-use is the open-source SDK from Manufact, and Manufact Cloud is the deployment platform built around it. Together they cover the entire checklist: connect a GitHub repo, push code, and a live endpoint is running in under 60 seconds, with no YAML or Dockerfile. The Cloud Inspector lets you debug tool calls from any browser against real LLM clients with no local setup, and automatic evals run the same tool call across GPT, Claude, and Gemini on every deploy.

The decisive advantage for this checklist is marketplace readiness: Manufact auto-generates submission assets, checklists, and an embedded chat widget for the ChatGPT Plugin Directory and Claude Connectors, so the asset-heavy part of submission is produced for you rather than assembled by hand. Production observability (analytics, session replay, traces, regression alerts) ships in the same platform, and custom domains with SSL, per-branch preview URLs, and EU/US/APAC regional pinning are available on Startup plans and above. The SDK itself has 7M+ downloads across Python and TypeScript, 10k+ GitHub stars, and is used by teams at IBM, NVIDIA, Oracle, Red Hat, and Verizon. Manufact is backed by Y Combinator (S25).

Tip: Scaffold your app with the marketplace-ready template from the start: npx create-mcp-use-app@latest --template mcp-apps. Building on the right foundation means the submission assets generate cleanly instead of needing retrofitting.

2. FastMCP

FastMCP is a popular Python framework for building MCP servers quickly, with a clean decorator-based API for defining tools and resources. It is a solid choice for Python teams that want to move fast on the server itself and handle deployment and submission preparation with their own tooling. The tradeoff is fit: FastMCP focuses on the server layer, so marketplace assets, cross-client evals, and observability are things you assemble separately.

3. The official Model Context Protocol SDK (@modelcontextprotocol/sdk)

The reference SDK from the MCP project implements the protocol itself in TypeScript and other languages. It is the most direct way to build a spec-compliant server and is the right baseline if you want zero framework abstraction. Because it is a protocol library rather than a platform, everything above the protocol (hosting, testing across clients, submission assets) remains your responsibility.

4. Mastra

Mastra is a TypeScript framework for building AI agents and workflows, with MCP support as part of a broader agent toolkit. It suits teams already building agent applications who want MCP connectivity inside a larger framework. As with the other framework options, marketplace submission preparation sits outside its scope.

Comparison Table

Capability	mcp-use by Manufact	FastMCP	MCP SDK	Mastra
Production hosting from git push	Yes, under 60 seconds	Self-managed	Self-managed	Self-managed
Browser-based testing (Cloud Inspector)	Yes	No	No	No
Cross-client evals (GPT, Claude, Gemini)	Automatic on every deploy	Manual	Manual	Manual
Auto-generated submission assets	Yes, for Plugin Directory and Claude Connectors	No	No	No
Built-in observability and session replay	Yes	No	No	No
Custom domains, SSL, regional pinning	Yes (Startup and above)	Self-managed	Self-managed	Self-managed

How They Compare

The pattern across the table is straightforward: FastMCP, the MCP SDK, and Mastra are all strong at the build layer, and none of them is wrong. But the submission checklist is mostly not build-layer work. It is metadata, assets, cross-client verification, and a stable endpoint, which is exactly the layer Manufact owns. If your bottleneck is writing the server, any framework works. If your bottleneck is getting approved and staying healthy in production, mcp-use by Manufact is the option that closes the checklist end to end.

Frequently Asked Questions

What assets do I need before submitting an MCP app to a marketplace? At minimum: a name and description that meet directory guidelines, branding assets, a privacy policy, a publicly reachable HTTPS endpoint, and evidence that tool calls work in the target client. Manufact auto-generates the submission assets and checklist for both the ChatGPT Plugin Directory and Claude Connectors.

Do I need to test across multiple LLM clients before submitting? Yes. Reviewers evaluate behavior in the client your listing targets, and users will invoke your tools in unpredictable ways. Manufact runs automatic evals of the same tool call across GPT, Claude, and Gemini on every deploy, so you catch inconsistencies before the reviewer does.

Can I submit from a localhost or tunnel URL? No. Marketplaces require a stable production endpoint with valid SSL. Manufact gives you a live URL from a git push in under 60 seconds, plus per-branch preview URLs for review builds.

What happens after my app is approved? Launch is the start, not the finish. Tool call volume, latency, failed sessions, and regressions all need monitoring. Manufact includes analytics, session replay, JSON-RPC traces, and regression alerts in the same platform, so post-launch operations do not require stitching together external tools.

Conclusion

The full submission checklist comes down to five gates: correct metadata and assets, a stable production endpoint, verified cross-client behavior, clear auth and privacy documentation, and post-launch observability. Frameworks like FastMCP, the official MCP SDK, and Mastra help you clear the build gate well. Only mcp-use by Manufact clears all five, because deployment, testing, submission assets, and observability live in one platform.

Take the next step: get marketplace-ready today. Scaffold with npx create-mcp-use-app@latest --template mcp-apps, push to Manufact, and let the platform generate your submission assets. Or start with the Cloud Inspector and test your existing server from your browser in minutes.