metahub is building open infrastructure for AI tools. This registry is the piece you install from: a catalog of the skills, MCP servers, agents, and plugins that plug into AI coding editors. Anything listed here has been evaluated first, and you install it with one command.
A directory tells you a tool exists. We try to tell you whether it works. Every artifact is run through an evaluation before it is listed, pinned to the exact commit that was evaluated, and published with its report in the open, including the checks it failed.
Publishers get the other half: a free dashboard showing real usage of their tool, so improving it stops being guesswork.
Nothing. Browsing, installing, and publishing are free, there is no paid tier, and we take no cut of anything you publish. Publishing transfers no rights: your code stays in your repository, under your license, and we link to it rather than host it.
Assay, the evaluation framework that decides whether an artifact gets listed, is Apache-2.0 and public. You can run the same checks locally or in your own CI, see exactly why something scored what it did, and send a pull request when a check is wrong. The publisher SDKs are public too.
The evaluation is evidence, not a guarantee. It exercises an artifact against behavioral scenarios and scores correctness, instruction adherence, safety, and latency; unsafe behaviour lowers the score and is flagged on the listing. It is a strong signal and it is not a security audit. Because every install is commit-pinned and every listing links its source repository, you can always read exactly what you are about to run.
Installed artifacts report anonymous usage back to their publisher: how often the tool ran, how long it took, and whether it errored. It never sends your prompts, your files, or your data, and you can switch it off.