MCP authentication monitoring: catch auth drift before customers do
MCP authentication breaks quietly. An issuer changes, a metadata field disappears, a client secret lapses — and uptime monitors stay green while every customer agent stops connecting. MCPComp scans your servers daily and alerts on exactly what changed.
mcpcomp https://mcp.example.com/mcp --baseline baseline.jsonThe same diff engine runs in the hosted dashboard on a daily schedule — add a server and the first scan becomes its baseline.
Why uptime monitoring misses auth failures
An MCP server's metadata stays perfectly valid right up to the moment its credential expires or its authorization server ships a change. The server still answers 200. The OAuth documents still parse. What broke is a relationship between documents — an issuer that no longer matches its discovery URL, a code_challenge_methods_supported that vanished — and no HTTP prober looks at that.
What monitoring gives you
A diff engine over the facts that matter
Each scan is compared against a committed baseline — the fields a verdict depends on, not raw responses, so a timestamp change is not drift but a changed issuer is. The changed fields arrive in the alert.
Alerts where you already work
On drift you get an email and — through the MCPComp GitHub App — a labelled issue in your repository plus a pull request committing the post-drift report, so accepting a change is a reviewable diff and merging it closes the issue.
The failure no probe can see
Expiring credentials never show up in public metadata. The Entra integration reads credential expiry from inside the identity provider and can rotate an expiring secret on request with zero downtime — both secrets stay valid during the switchover.
Frequently asked questions
What counts as drift?
A change in any fact a verdict depends on: issuer identity, advertised PKCE methods, registration support, endpoint TLS, grade-relevant metadata fields. Timestamps and transient HTTP details are deliberately excluded.
How do alerts arrive?
Email, plus a GitHub issue and a baseline-acceptance pull request in the repository you bind through the GitHub App. Merging the PR accepts the new baseline and closes the issue.
How many servers can I monitor?
One on the free plan, ten on Pro ($30/seat/mo billed annually). Custom limits on Enterprise.