Manufact is the MCP Cloud to get ready for every marketplace, monitor with MCP-first analytics, and run evals against every harness and client.
We're the team behind mcp-use, the open-source MCP framework with 10k GitHub stars and ~10M+ downloads.
This is the 2nd launch from Manufact. View more

Manufact
Launching today
Manufact helps teams publish MCP servers and plugins, understand how agents use them, and test releases across clients. Bring publishing checks, session traces and cross-client tests into one workflow.






Free Options
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted




Hi Hunters 🐱👋
I’m Luigi cofounder of Manufact (YC S25) the MCP Cloud, to get ready for every marketplace, monitor with MCP-first analytics, and run evals against every harness and client.
Deploy, test, publish and monitor your MCP servers and plugins, all in one place.
Teams like AgentMail, Alpha Vantage and Context.dev use Manufact.
And a fun full-circle moment: even Product Hunt used our tooling for its own MCP!
🚨 The problem we’re solving
Getting your MCP working is only the beginning.
Does it work inside the clients your users actually use?
Is it ready for marketplace review?
When an agent fails to complete a task, can you find out why?
We built Manufact to answer those questions.
🥁 What you get
1️⃣ Deploy
Connect your GitHub repo and get a live MCP endpoint. Merges to main deploy automatically, and every PR gets a preview deployment so you can test changes before shipping. Logs are right there when you need them.
2️⃣ Test
Test the whole plugin lifecycle, wherever you host your MCP.
Run evals across models, harnesses and operating systems, including real ChatGPT and Claude clients. An LLM judge grades each run, and you can inspect every tool call to understand what happened.
Go beyond “the tool returned a response” to “the user actually completed their task.”
3️⃣ Publish
Prepare and manage your marketplace submissions from Manufact.
Run publishing checks, get a readiness score and use Auto fix to address issues. Then generate your submission pack: listing materials, reviewer test cases and supporting assets in one ZIP.
4️⃣ Monitor
Get analytics built for MCP.
Follow sessions from users and agents, see which tools they call, and track latency and errors. When something breaks, open the session and inspect the calls behind it.
🏰 Our story
We started mcp-use SDK (10k github stars) because we wanted building with MCP to be easier.
As developers took their projects into production, the next problems became clear: deployment, client differences, marketplace requirements and understanding real usage.
Manufact brings that work together, so you can spend more time building your product.
Think Vercel for MCP, with testing, publishing and analytics built in.
🚀 Special Product Hunt launch offer
The first 100 signups can use PH100 to get $100 in Manufact credits.
Try it here 👉 https://manufact.com/producthunt
The offer is available whether or not you engage with this launch.
What’s been the hardest part of taking your MCP from a working demo to production?
Drop your questions below.
We’re here and would ❤️ your honest feedback!
This is by far the best infra for MCP.
We used it to deploy an app that interfaces with ChatGPT - and our users love the app :)
Thanks @zuri_obozuwa! Looking forward to see you soon at another hackathon building other cool creative stuff based on MCP!
Congrats on the launch team!
Thanks @charlotteschmitt!!
Best MCP infra! Congrats on the launch :
@vicdevelopmently thanks Vic, big brain!!
ok this one hits close, an MCP that passes every local test and then picks the wrong tool inside Claude is a whole diferent bug
we run one for mailX, so on your question: client differences by far. when something works in Claude but fails in ChatGPT, does the judge tell you why or just that the task failed??? and is the readiness score per marketplace???
@thamibenjelloun thanks for the kind words! yes, our tests not only will give you a success score, a full transcript of the conversation, and all the tool calls with their arguments but also will signal the specific problem that made the task fail! The readiness score is per marketplace, you will see if your MCP is ready for some but not for others, so you can also submit gradually!