I Set a 500ms Timeout and Got 28 Hours

A MCP server was experiencing intermittent hangs, leading the reporter to implement a 500 millisecond timeout. This timeout value, specified in milliseconds, was set to a minimum, effectively limiting the server's response time. However, the reporter later discovered that values below 1000 milliseconds were ignored, and instead, the system defaulted to a 28-hour timeout.

This discrepancy led to confusion, as the reported 500 millisecond timeout was, in reality, a 28-hour timeout. The reporter also learned that progress notifications from the server did not affect the timeout, and a timeout of at least 1000 milliseconds served as a floor on the idle timeout. Additionally, the reporter found that URLs without a specified type were treated as stdio servers, requiring additional configuration.

A common issue the reporter encountered was a repository unable to approve its own servers, which appeared as a permissions problem. This occurred because servers from .mcp.json files needed approval, which could only be granted through specific settings files. The reporter emphasized the importance of a configuration checker that doesn't flag correct configurations, as the existing checker would overlook valid settings and allow unintended behavior.