From 31 Aug 21:58 UTC to 2 Sep 14:39 UTC this page showed the MCP Server as down. That was a monitoring misconfiguration, not an outage: after a routine server update the endpoint started answering the health probe with a different, still valid, response code, and the check treated it as a failure. The MCP Server remained available throughout that period. The check has been replaced with one that accepts both responses.

Separately, the MCP Server was genuinely unavailable on 2 Sep between 13:25 and 14:19 UTC while a deployment was being recovered. That incident is resolved.