When Marcelo Jimenez Rocabado decided to fork Anthropic's open-source MCP Inspector in late 2024, he wasn't just starting another developer tool. He was making a bet that one of the most well-funded AI companies in the world had fundamentally misunderstood what developers needed to build the future of AI agents. Six months later, the 24-year-old CTO of MCPJam has proven his instincts right. The project has exploded from 200 to over 1,300 GitHub stars, built a thriving community of 200+ Discord members, and attracted venture backing from Open Core Ventures, all while Anthropic's original Inspector languishes weeks behind the MCP specification with no clear roadmap. MCPJam "They didn't hire a team for this," Marcelo says, reflecting on why Anthropic's approach fell short. "There are only two full-time maintainers for a project that gets a lot of issues, a lot of PRs per day. It’s more than two people to handle." The problem Anthropic couldn't solve The problem Anthropic couldn't solve The Model Context Protocol, released by Anthropic in November 2024, fundamentally shifts how AI agents interact with tools and data. Companies like Asana, PayPal, Vercel, and GitHub are already building MCP servers to prepare for the agentic transformation. The problem is that no one has built true testing infrastructure for these servers. Marcelo discovered this firsthand while building MCP servers himself. The Anthropic Inspector existed, but it was incomplete. "It did the job, but it could have made better improvements," he explains. "The developer experience lacked features. It was open source, but hard to contribute to. And once you did, it also required some time to review." The breaking point came when Marcelo realized Anthropic wasn't just slow. They were philosophically opposed to what developers actually needed. "They don’t do LLM testing and all-indeterministic testing. It’s kind of counterintuitive given that MCP is made for LLMs," he says. While developers were begging for an LLM playground to test their servers against real language models in real environments, Anthropic never built it. So Marcelo, along with his co-founder Matthew Wang, did instead. A contrarian bet on MCP's future A contrarian bet on MCP's future While most developers in the community expect clients to improve, Marcelo believes the real opportunity lies in building better servers. "I think a lot of people are expecting clients to be better," he explains. "But at the end of the day, I think servers have to be better and don't rely on the client because right now a lot of developers only use three clients, which are like Cursor, Claude, and ChatGPT." As the AI agent ecosystem matures, there will be hundreds or thousands of different agent clients deployed. "You can't be assuming that they have the same system prompt as these other big ones, or they're not going to have the same tools, like a web browser tool, for example. So you just want to build a server that's defensible against any client." This insight stems from his broader bet on the future of AI. "I think everything's going to be agent-run, and we're making a bet in the layer that's going to enable that," he says. "It's not even going to be about what model it is. It's going to be more of, like, who's managing the context better." For Marcelo, MCP isn't just a protocol. It's the protocol. "It's just in the name, right? Model Context Protocol. It's all about that." Building with a community focus Building with a community focus Since forking the Anthropic Inspector, Marcelo and his co-founder Matthew have moved with remarkable speed. Under the guidance of Open Core Ventures' Catalyst program, they've not only grown their GitHub stars but built genuine community infrastructure. "If we didn't join OCV, we would not have built our Discord community. We would not have posted good first issues on GitHub, which led to new contributors supporting the project," Matthew admits. That shift from pure user acquisition to community building has paid off. MCPJam now has seven active external contributors. The product roadmap is equally ambitious. While the open-source MCPJam Inspector provides core testing and debugging capabilities, the team is developing a premium MCP evals framework that will benchmark server performance across all environments and LLMs. "Build that into your CI/CD pipeline, and we can monitor whether changes to the MCP server made an improvement or introduced a regression," Matthew explains. But perhaps most tellingly, major companies are already taking notice. The team is now testing MCP servers for enterprise clients and helping them improve their server design. As the enterprise world accelerates toward an agentic future, Marcelo is boldly betting that every company with an API will also build an MCP server. The question isn't whether Anthropic made a mistake in under-investing in their Inspector. The question is whether they can catch up to the 24-year-old who saw the gap they left behind. For Marcelo, the answer is clear. "We are an MCP-first company. We can move faster, and we have more domain expertise," he says. "If an MCP server works on MCPJam, it'll work anywhere." With over a thousand developers already betting on his vision, Marcelo is proving that sometimes the best response to a tech giant's oversight is to build what they couldn't, or wouldn't.
