Stars and use

GitHub stars show interest. They do not show who set a tool up. Here are 90 MCP server packages with both numbers: the stars on the package's GitHub repo, and how many GitHub accounts have it in a public repo's agent config now. Across all of them the rank correlation between the two is 0.45, where 1 would mean stars and use rank the tools the same way.

Used more than starred

The most accounts per star.

Starred more than used

The fewest accounts per star, among repos that are about the MCP server itself.

How this is counted

Only MCP servers that come as an npm or PyPI package, with a tool page here and 20 or more accounts. The repo is the one the package's own registry entry names; none was picked by hand. An account counts once, however many of its repos have the tool. When several packages share a repo, the repo is listed once with its most used package. The rules were written down before any star count was fetched.

What this can't tell you

Low use here does not mean a tool is bad. Favz sees config checked in to public repos, not what people run from their own machine, and many servers are set up that way. Stars also pile up over years, while most of this config is recent. A repo with many packages gets one row, so its stars are compared with one package.

Known faults. The repo comes from the package's registry entry, and that can name a small side repo: the "prisma" package points at prisma/prisma-cli, not the main Prisma repo, so its stars here are far too low. A short package name can also be typed by people who meant a better known tool with a similar name. Read the top of the first list with that in mind.

All 90 rows are in index.json.