We implemented a fix and are currently monitoring the result.
This is only affecting the hosted MCP server for the API, a bad deploy has caused all requests to fail that are sent to it.