I always felt lots of stars was a signal something was half decent, but wanted to search a bit deeper and decided to analyze the followers of some big-ish popular AI repos
So I analyzed stargazer data (from public profiles) for some MCP repos and bionic-gpt (self hosted chat gpt alternative)*
*This is not a statistically perfect study. It’s more “I wrote some scripts, squinted at the plots, and a few patterns were too interesting not to share.”
1. MCP: boring use cases are starting to win
I still thought MCP was a kind of untested new thing that some people plugged into their vibecoding workflows, but if found there’s a lot of stargazers with big company jobs, vendors integrators larger companies and the more I looked into it i found there was significant traction in the ERP accounting, financial space already with Microsoft Dynamics etc.
2. Metacognition & “slow” architectures are a big deal
The second thing I noticed was looking at what other repos people starring mcp were interested in and I came across other repos and blog posts about using low powered models with loads of steps stuff: tiny recursive models, meta-controllers, agent planners, tool frameworks.
Tiny models with loops can beat giant models on reasoning-heavy benchmarks.
Plain LLM + a feedback loop often matches more expensive “reasoning” models.
3. Self hosted chat-gpt: big in europe & china
Only ~23% of stargazers of bionic-gpt list the U.S. as their location, most are in Germany, China, France, UK, Canada, India, without europe significantly outweighing the US. A couple of people said “Mars” too - not sure what to take from that.
That lines up almost too neatly with:
Europe caring about data locality / GDPR / on-prem
China not having normal access to OpenAI and leaning on local or open alternatives, but the chatgpt style ui and format is universally valuable
So… what’s the point of all this?
Mostly: who stars you is often more interesting than how many. If you maintain an OSS project, running this kind of analysis even once can give you a better idea of your actual audience than “star count went up.”
Tiny tool footnote
Doing this manually with browser tabs and spreadsheets got old fast, so I hacked together a small tool that takes a GitHub repo URL → returns some aggregate stats about its stargazers
(roles, seniority, top languages/frameworks, companies, locations, etc., using public profile data)
It lives here
I built it mostly for my own curiosity, so it definitely has rough edges. If anyone wants to poke at their own repo(s), I’m happy to throw as many free credits your way as you need to test it, I mostly want feedback on whether this sort of analysis is actually useful for other people or if I’ve just over-optimized my hobby.
Top comments (0)