When “session not found” is really a deployment bug
What jumped out at me is how ordinary this failure sounds once you strip away the MCP jargon: a stateful protocol gets thrown behind a load balancer and then everyone acts surprised when memory disappears. That part is not subtle. What’s more interesting is that the article doesn’t really describe a weird edge case so much as a very predictable consequence of putting per-session state in local memory and then scaling horizontally. I’m mostly convinced by the diagnosis. If the session lives on on
papoo.work