Use tashan
Start with the stack you already have.
Most advice tells you what to install. The more useful question is what you're already running that shouldn't be there — a server the author archived, a package marked deprecated, or one whose name shadows an official one. One command answers it.
1 · Audit your config
nothing is uploadednpx tashan-cli doctorReads the agent config on this machine — Claude Code, Claude Desktop, Cursor,
VS Code, Windsurf, and ~/.claude/skills/ — and matches every entry against the Index.
It prints only what needs a decision. A real machine has a hundred-odd skills
that are simply fine, and listing each one buries the two servers that matter — so those are
counted, and --all lists every row.
- ! deprecated on npm, archived repo, removed from the registry, or a name that shadows an official package
- ~ no recent activity — looks abandoned
- · catalogued but not rated — we know it exists and nothing about its quality
- ? not in the Index at all — unmeasured, which is not the same as bad
doctor reads files
on your disk and compares them against a public JSON file. There is no account, no key, no beacon and no
upload. If we ever add aggregate contribution it will be a separate, explicit command — and the rule we
hold ourselves to is that any data that feeds a score must be published. A score computed from
numbers only we can see is exactly the thing this site exists to be an alternative to.Exit code is 2 when something needs attention, so it drops into CI:
npx tashan-cli doctor || echo "stack needs a look". Add --json to pipe it.
2 · Find something worth installing
npx tashan-cli search database
npx tashan-cli top browser
npx tashan-cli info context7Or browse in the page: the full Index, filtered by type, activity and depth — or jump straight to a category such as Database, Browser & Web or Dev Tools & CI. Every row links to a dossier showing the arithmetic behind its score.
3 · Install it
npx tashan-cli add context7 --client claudePrints the exact command or config block for claude, cursor,
desktop, codex, or plain npx — generated from the same logic as the
website, so the terminal and the page never disagree.
For maintainers
If you publish a capability, embed its measurement. The badge is generated from the same data as the Index and updates when the evidence does — including downward. That is the point of it.
Your slug is on your capability page. Nothing about a badge, a claim or a payment changes a score — see the methodology.
Let your agent ask before it installs
Everything above assumes you are the one typing. If you would rather your coding agent check before it recommends or installs anything, install tashan into it once:
claude mcp add tashan -- npx -y tashan-cli mcpThe agent then gets three tools — find a capability for a job, check one before installing it, and audit the config already on the machine. All three are free; nothing about risk is ever behind the paid tier.
For agents and machines
/llms.txt— the site as plain markdown: how the score works, the top capabilities, every category. Written to be read by a model, not rendered./data/capabilities.json— the full export, every field, no key required.npx tashan-cli search <q> --json— the same data through the CLI, for scripting.
Scores change as the evidence changes, so cite the date. Attribution: tashan (tashan.sh), measured <date>.