Author here. I build MCP Canary, a monitor for MCP endpoints, and the probe behind this is its inspect step (official C# SDK, Streamable HTTP with SSE fallback, no credentials, never calls a tool)
Things that not surprised me with all AI slop: listings not updated since April are broken 40% of the time; Railway and Render endpoints are broken 50% / 39% vs 6% on Workers; 78% of servers behind auth don't declare a required header in the registry
Things that not surprised me with all AI slop: listings not updated since April are broken 40% of the time; Railway and Render endpoints are broken 50% / 39% vs 6% on Workers; 78% of servers behind auth don't declare a required header in the registry