MCP servers & agent skills · measured, not claimed
Find what's worth installing for your actual work
Audited first — known advisories, what runs at install time, and what it can reach on your machine.
Start here
What do you do?
Pick the work and the board below re-ranks to the capabilities measured for it. Every count is capabilities on the board right now, not a catalogue total.
The Index
Ranked by the tashan score
| # | Capability | tashan | Evidence | Health |
|---|---|---|---|---|
Loading measured data… | ||||
Already running things
Is anything in my stack dead or vulnerable?
One command reads your Claude Code, Cursor and Desktop configs and flags known advisories, install-time scripts, and anything deprecated, archived or abandoned. Local only — nothing is uploaded.
npx tashan-cli doctor ›
You are an agent
Can my agent check before it installs?
A 50 KB lookup of every score, the full set in MCP-registry shape, and a SKILL.md you can install so checking is a tool call. An absent name means unmeasured, never bad.
curl tashan.sh/v0.1/scores ›
Why trust this
Ranked on evidence, not spend.
Earned, not bought
Adoption, upkeep, freshness, real expertise — every rank comes from public evidence. No stars, no sponsorships, no louder listing moving up the board.
We read the actual expertise
Directories list; we open each capability and grade the real thing — deep domain work versus a thin wrapper or AI-slop. GitHub evaluates code. Nobody else evaluates the expertise.
Every number, re-derivable
No black box. Each signal is defined and checkable against public sources — the registry, npm, git history. If you don't believe a number, verify it.