How tools get listed

Coding agents read this list to decide what to install, so a recommendation is only worth something if it can't be bought. Nobody can pay to be listed, to be described differently, to be named for a job, or to move up.

That covers everything an agent reads: llms.txt, tools.json, the agent skill, and the tool entries and “use when” notes on these pages. The featured tools on the homepage are editorial picks too.

If we ever run sponsorships on the website, each one will be labelled “Sponsored”, kept apart from the directory, and never included in anything agents read. Partnerships with tool makers (shared content, starter templates, best-practice guides they publish) don't change how their tools are listed.

A JavaScript or TypeScript library or tool that someone would choose for an AI job: calling a model, building an agent or workflow, MCP, retrieval and vector storage, running models in the browser, or coding with AI. It has to be:

• installable from npm (or, for dev tools, a product developers use), with a public source repository;
• maintained: not deprecated or archived by its maintainers, with recent releases or commits;
• used by real projects. Download counts help us notice a tool; they don't decide whether it fits.

Utilities, adapters and internals of a listed project are covered by that project's entry rather than listed separately.

Each entry says when to pick it over the others in its category, and the “choosing by job” lines in llms.txt name the best default for common jobs, with an alternative when the choice depends on what you need. We judge by fit for the job, how current and maintained the tool is, and what agents actually build with it in our tests.

Deprecated or replaced packages stay listed, marked as such, with the replacement named, so an agent that remembers the old name is pointed to the new one.

Every morning a check asks npm and GitHub about every listed package and repository: deprecations, archived repos, new major versions, new packages from listed projects, and popular AI packages we don't list.

Our maintenance agent then verifies each finding against npm and GitHub and updates the list. Updates publish on their own only when an automated check passes: the change is data only, removes nothing, and every new package is confirmed on npm and comes from the project's own repository. Anything else waits for a person.

Something wrong, out of date or missing? Suggest a tool or email us. Maintainers can send release notes or best-practice guides for their tool; we use them to keep its entry accurate, the same way for every project.