Don't install a dead MCP server.

Third-party audits found roughly half of public MCP servers abandoned — yet no registry shows you a pulse. Watchtower scans the ecosystem continuously and scores every server: ALIVE, FADING, or DEAD.

Healthiest servers

Top of the scanned set, ranked by health score. Score = push recency (70%) + community confidence (15%) + maintenance hygiene (15%).

#	SERVER	SCORE	STATUS	STARS	LAST PUSH	WHAT

The graveyard

Servers that are dead (no meaningful activity for months, or archived) or gone entirely (repository deleted). Installing these is how agents break silently.

SERVER	SCORE	STATUS	STARS	LAST PUSH

Health badge for your README

SHOW USERS YOUR SERVER IS ALIVE

If you maintain an MCP server, embed your live health badge. It re-renders on every scan — proof of maintenance that star counts can't fake.

EMBED

Markdown — replace owner/repo with your repository:

![MCP health](https://wocaoac-cpu.github.io/mcp-watchtower/data/badges/owner__repo.svg)

Badges ship with every scan in data/badges/.

Methodology

Push recency — up to 70 pts. The dominant death signal: audit medians show dead servers sit ~142 days since last commit. ≤14d = 70, decaying to 2 past a year.
Community confidence — up to 15 pts. Log-scaled stars. Popularity ≠ health, so it can never rescue a stale repo.
Maintenance hygiene — up to 15 pts. License present, issues enabled, description set, not a fork, issue backlog under control.
Archived → 3. Repository deleted → GONE.
Verdicts: ALIVE ≥ 70 · FADING 40–69 · DEAD < 40. Thresholds are public and debatable — open an issue.