Fleet status for MCP servers
Register your fleet, then pull a live dashboard that scrapes each server’s real
/health and /metrics endpoints on demand. Every number comes
from a live HTTP call you can re-run yourself — nothing estimated, nothing cached
past the last scrape.
Three endpoints. No background daemon, no agent to install on your servers.
Name and base URL, with optional custom health/metrics paths if a server doesn’t use the defaults.
Hits every registered server’s /health and /metrics right now, parses the
Prometheus exposition text with the standard prometheus_client.parser library — not a
custom regex — and returns real request/error counts and error rate per server.
Every scrape is logged, so you can see whether a server’s error rate is climbing across calls, not just its state at this instant.
This is the actual shape of GET /v1/dashboard against two registered servers
— one healthy, one throwing errors under load.
{ "servers": { "prod-mcp-1": { "status": "up", "total_requests": 15422.0, "error_requests": 18.0, "error_rate": 0.001167 }, "prod-mcp-2": { "status": "up", "total_requests": 6031.0, "error_requests": 742.0, "error_rate": 0.123032 } }, "total_servers": 2, "up": 2, "down": 0 }
Trend Micro found 492 MCP servers exposed to the open internet with zero authentication in 2026. That’s not a sophisticated-attack problem — it’s teams not having a simple answer to “which of our MCP servers are up right now, and which are throwing errors.” Generic APM (Datadog, Grafana, New Relic) already solves monitoring in general; MCPWatch solves this one specific, currently-unaddressed gap.
No distributed tracing, no log aggregation, no alerting/paging. Run Grafana/Prometheus/Datadog alongside MCPWatch if you need those — this is a focused fleet-status dashboard, not a replacement for a full observability stack.
On-demand scraping, not a running collector. GET /v1/dashboard
scrapes live when called. Wire it into your own cron or scheduler for periodic snapshots.
Your servers need to already expose /health and Prometheus-format
/metrics. MCPWatch doesn’t discover servers on your network —
you register them explicitly.
Runs in your own infrastructure. No per-seat fees, no usage metering on your data — you own the container.
docker run and you’re monitoring in minutes