The road to 10,000 stars #10000

mcp-use was born on March 28, 2025. The MCP protocol was released just a few months before and it was starting to gain traction. The first clients that supported MCP were Cursor and Claude Desktop, where you could connect your MCP server by pasting a JSON configuration file.

The most popular MCP server at the time was Playwright, that gave agents access to Chrome. Other popular MCP servers were all development oriented, docs, GitHub etc. We immediately saw the bigger picture. At the time MCP was a developer only product, but the ability to give agents access to external systems by just plugging in an external component could have much bigger impact.

If this was going to be true, proprietary closed source applications could not be the only way to use MCP servers.

So we wrote mcp-use, it was a simple library that allowed you to connect any MCP server to any LLM in six lines of code. Fun fact: initially the library was called pymcp, then mcpeer, it landed on mcp-use just after (inspired by our friends at browser-use).

As every open source project, mcp-use started with a README file and relentless posting on Reddit. Demo after demo, we started collecting some interest and feedback. The first demo that went "viral" was the one that showed how to control a curtain with MCP. It showed that MCP is much more than a way to connect your coding agents to documentation.

The initial demos are always hard. The product is so simple that it seems like it is not worth launching. The first updates are so small that it seems hard to think that anybody will care.

Fortunately, we did not stop, and at some point LangChain picked up on it and decided to shout it out, giving massive visibility. The post went viral and so our repo.

After the initial developer love and traction from our open source community, we certainly were convinced that we had something going on. In addition we strongly believed in the cause and we loved working on mcp-use, we could see ourselves working on it for the next 10 years, and we decided to make this our life work.

We applied to YC S25.

YC not only is the best place to build any startup, But recently had started funding some of the best open source dev tool companies out there, Supabase, PostHog, Browser Use, just to name some. It seemed the perfect place to grow our open source project into a real company.

Both of us had applied to YC several times in the past. What was different this time? As many things in life, things happen when you have the least pressure. We were so motivated and excited about working on mcp-use, that in our mind YC was not a necessary condition for our success. Hadn't we got it, we would have continued anyway.

We submitted our application very spontaneously, with full transparency and confidence in what we had built.

A couple weeks later, while Luigi was in boarding line of a plane, we had the final call with our Partner Jon, who gave us the good news.

We were accepted to the YC S25 batch.

YC was the perfect place for us to grow, the relentless pace of launches, demos, customer calls helped us iterate faster than ever.

We launched our first cloud product. Talked to hundreds of companies, startups and builders. Launched new features. Every time we made something happened, we put it out there. During the YC batch we launched around 10 times, we let as many people as possible know what we were building. In just 6 weeks we doubled our stars, tripled our downloads, closed our first customers. Our SDK was used by 20% of the US 500.

The lesson: make sure people know about what you are building. It will drive your growth or tell you that you are doing something wrong.

At the end of the YC batch, we have proven our thesis. We have a strong community of developers using our open source library and we have a cloud product used by customers in production.

We decided to raise a seed round to fund the growth of the company and the development of the open source library.

We raised from some of the best investors out there, investors and operators that know dev tools and open source.

This was extremely important for us. We want partners that understand the open source world, the value and importance of the community and will help us grow the project in the right way.

We could not be luckier in this sense.

After YC, we were in full open-source build mode.

We wanted our SDKs to provide the best developer experience accross the MCP stack.

So far, mcp-use was mainly helping developers consume MCP servers, through our MCPClient and MCPAgent abstractions. To build good clients, you also need servers. While building mcp-use we have tested, built and debugged many MCP servers. We had so many ideas and feedback from the community that we decided to build our own server SDKs.

That is what we did!

We started working on our server SDKs, we released a TypeScript version of it, and an open source inspector to test and debug MCPs just like web apps.

In February, we hosted the largest MCP Apps hackathon ever at YC headquarters in San Francisco.

We had 650 applications, 300 builders, sponsored by OpenAI, Cloudflare and Anthropic. Prizes included tens of thousands of dollars in credits, Mac Minis, and a YC interview!

Such a great experience, being in the same room with so many builders, the panel of judges who are writing the MCP spec, was a great source of feedback and inspiration for us.

It is also where we coined the term "Make something agents want" which kind of became our mission statement.

We were Platinum Sponsors at the MCP Dev Summit in New York City. Another occasion to meet with the MCP community, steering committee and most talented and forward looking companies in the space.

Right around the MCP Dev Summit time, we welcomed @khandrew1@khandrew1 to the team!

The release of MCP Apps as an official extension, the opening of the ChatGPT and Claude MCP Stores, was a huge moment for MCP, and we wanted to be there to support it. We have been working hard on our products to make it super easy to deploy, monitor and publish production MCP Apps.

The MCP App hackathon, the MCP Dev Summit and many conversations we had with companies and builders that wanted to publish their MCP apps, gave us the right directions.

We listened. Improved our SDKs and product to serve this new wave and made manufact.com the best place to build, deploy and publish MCP apps and servers.

If you do not want to take my word for it, take OpenAI's:

We launched all the features in a launch week, with a new feature every day. Check it out here.

A star is one developer deciding this was worth remembering. Ten thousand of those, mostly from people we never met.

Thank you to everyone who starred, filed an issue, opened a PR, or shipped a server.

What is next?

We are keeping our heads down, iterating on our product and feedback, making sure our SDKs provide a first class experience, following the latest trends in the SPEC.

We believe MCP is going to cause a massive change in how software is built and distributed, and we want to power this change.

Our SDKs to build, our Inspector to test, and our Cloud to deploy.