@samanhappyMCPHub and Selectly: open source from a Nanjing engineer
Hi everyone, I'm Saman Sun, a software engineer in Nanjing. By day I work on enterprise AI systems. Outside of work I maintain MCPHub, an open-source, self-hosted MCP gateway written in TypeScript that lets you connect, manage and govern MCP servers in one place. I also build Selectly, a browser extension that lets you select text on any page and translate, explain or polish it with AI. I use Codex for most of my development.
MCPHub has grown to about 2,500 stars on GitHub, and I'm now focused on making it a gateway people can trust in real setups. I'm also working out how to move from engineer to independent product builder.
What I'd love from this community: an intro to anyone running MCP servers in production, or to someone who has made the jump to independent products and is willing to share what they learned.
Comments2
Welcome to Lounge, Saman. For an MCP gateway to be trusted in real setups, I'd guess per-server auth and audit trails matter as much as routing. Which governance gap are you hearing about most from MCPHub users running it beyond a hobby setup?
Thanks, Ryan! Great question. My current take is that fine-grained access control is the biggest gap — who can access which MCP servers and tools, especially as teams move beyond individual setups. Auditability is closely related. I'm still validating these needs with real users rather than assuming what production teams want. Have you seen similar challenges in practice?
2 comments
Welcome to Lounge, Saman. For an MCP gateway to be trusted in real setups, I'd guess per-server auth and audit trails matter as much as routing. Which governance gap are you hearing about most from MCPHub users running it beyond a hobby setup?
Thanks, Ryan! Great question. My current take is that fine-grained access control is the biggest gap — who can access which MCP servers and tools, especially as teams move beyond individual setups. Auditability is closely related. I'm still validating these needs with real users rather than assuming what production teams want. Have you seen similar challenges in practice?