Master MCP Server

# Performance & Tuning Tune routing and runtime to handle your workload. ## Routing - Prefer `health` strategy to route to instances with better health scores (when provided) - Increase `retry.maxRetries` for flaky networks, but cap `maxDelayMs` - Use `jitter: full` to avoid thundering herds - Adjust `circuitBreaker` thresholds based on observed upstream reliability ## Node Runtime - Use `NODE_ENV=production` - Run behind a reverse proxy (nginx, Cloudflare) for TLS termination - Set `LOG_LEVEL=info` or `warn` ## Workers Runtime - Bind KV storage for token persistence to avoid in-memory losses across isolates - Avoid large responses; stream when possible ## Observability - Use `/metrics` to scrape basic system stats in Node - Add platform logs/metrics (Cloudflare, Koyeb dashboards)