The corpus
What the first full pass found
17,317 public endpoints polled
49.2% healthy
39.9% behind auth
9.0% dead
39.7% of healthy servers on a protocol revision older than 2025-11-25
38 unrelated hosts serving a byte-identical tool list
Measured 2026-09-03 across every public remote MCP endpoint we could find. Full numbers in the launch post next week.
Free CLI
Try it now
npx mcp-drift-check https://your-server.example/mcp- Health category. The same handshake a real client does — legacy
initializeor modernserver/discover, Streamable HTTP or HTTP+SSE — resolved to one of seven states. - Tool list. Every tool the server exposes, with its description, as your agent would see it.
- Drift since your last run, with severity: a removed tool or a changed
inputSchemais breaking, a changed description is a warning.
Runs locally, stores snapshots in ~/.mcp-drift-check, and sends an anonymous usage
ping (version, category, exit code, platform — never the URL). Opt out with --no-telemetry.
Coming in 2–3 weeks
What the hosted monitor does
- Hourly checks of every endpoint you watch; every 15 minutes on paid plans.
- Tool-schema diff with severity. A removed tool or a changed
inputSchemais breaking; a changed description is a warning, because descriptions steer the model's tool choice. - Alerts by email, Slack, and Discord webhook.
- A public status badge for server authors.
- Seven days of history on the free plan.
Pricing
Three plans
| Free | Pro $19/mo | Team $49/mo |
|---|---|---|
| 3 endpoints | 25 endpoints | 100 endpoints |
| Daily checks | Hourly checks | Every 15 minutes |
| 7 days of history | 30 days of history | 30 days of history |
| Email alerts | Email alerts | Slack and Discord webhooks, status badge |
Prices are a plan, not a promise. Waitlist members get the first month free.
Waitlist
Get told when it opens
One email at launch. No drip sequence.