The idea is this: as a user, I might want to run a large amount of MCP servers, which could have a significant memory footprint or even some negative impact on CPU usage/battery life on laptops.

To mitigate this, we could have a serverless/start-container-on-demand mode where a proxy server is always ready to take requests on behalf of an MCP, but the actual container is not always continuously running. Instead, we would start the container on demand, and keep it running for a while, perhaps 30 min. After it's been idle for a period of time, the container can be stopped gracefully, but the proxy server keeps receiving requests, so the MCP will appear to be continuously online.