Comparison: Before vs After

MetricWithout MuxWith Mux
Tools in AI context50+ (all servers)4 (fixed)
Credit cost per request (Kiro)100%~5% (95% reduction)
Token cost per request (API)~10K tokens~600 tokens
Running processes15+1 + on-demand
RAM usage (idle)~800MB+~50MB
OAuth re-auth frequencyEvery enable/disableOnce (then cached)
Add new serverEdit client config + restartEdit servers.json (hot-reload)
Cold start to first tool callN/A (always running)1-3s (server spawn)
Subsequent calls (warm)Instant<100ms overhead