5e9f798 · 2026-09-10 19:30Z
System status

Measured, not promised

Every result on this page was produced by a request made when you loaded it. Nothing here is cached, scheduled or assumed, and anything that could not be measured says so instead of showing a colour.

Something is not fully healthy

At least one checked component answered, but not in the way it should — or a component could not be checked at all. The rows below say which, and how it was tested.

Checked 12 Oct 2026, 05:29:03 UTC

Components

  • Web app

    Not probed. This page was server-rendered by the web app to answer your request, which is itself the evidence.

    Inferred from this page rendering
    Operational

  • REST API

    Answered 200.

    GET /api/health
    HTTP 200146 msOperational

  • MCP endpoint

    Challenged the unauthenticated request with 401. That is the expected healthy answer for a Bearer-only transport.

    POST /api/mcp/mcp (unauthenticated)
    HTTP 401138 msOperational

  • Database

    Supabase Postgres is not checked here. A database failure would usually surface as a failing REST API check above, but that is an inference, not a measurement.

    Not probed from this page
    Unknown

  • Hosting platform

    Vercel and Railway platform state is not checked here. Reporting a provider's health without querying that provider would be a guess.

    Not probed from this page
    Unknown

How these checks work

  • Probes are aimed at https://viralzbackend-production.up.railway.app resolved from API_ORIGIN.
  • A 401 from the MCP endpoint is a healthy result, not an outage. The transport is Bearer-only, so an unauthenticated request is supposed to be challenged — receiving the challenge is what proves the route is mounted and enforcing auth.
  • When a probe gets no HTTP response at all — a timeout, a refused connection, a DNS failure — the component is reported unknown rather than down. From inside a serverless function, a blocked egress and a dead upstream look identical, and claiming an outage we cannot prove is no better than claiming health we cannot prove.
  • There is no uptime percentage and no incident history on this page, because there is nowhere to store either. A 90-day figure with no 90 days of data behind it is a fabrication.
  • Each probe is given six seconds. A component that answers slower than three seconds is reported degraded rather than operational.