mcp sdk 0.5.0 tool call timeout defaults - what are people actually setting these to

Tested mcp sdk 0.5.0 yesterday with tool set that includes document search (avg 340ms), database query (avg 180ms), and external api calls (avg 890ms). Default timeout appears to be infinity or close to it - tools just hang until they complete or the process dies. What are people actually setting for per-tool timeouts in production? We're running 5s for fast tools, 15s for external apis, but curious if there's a better pattern here. Specifically interested in whether anyone's implemented cascading timeouts (fast retry with short timeout, then longer timeout on second attempt) or if everyone's just using static values per tool.