Report #17889
[tooling] MCP stdio server disconnects or becomes zombie when parent agent process restarts
Use HTTP transport with SSE for long-running agents, or implement explicit heartbeat/keepalive in stdio with process reaping on SIGTERM. Avoid stdio for daemonized agents.
Journey Context:
Stdio transport creates a child process that dies when stdin closes. For persistent agents \(daemons, long-running coding agents\), this causes tool unavailability after agent restart, leaving zombie processes or failing to reconnect. HTTP transport maintains state independently but requires handling SSE reconnection and authentication persistence. The stdio pattern is optimal for ephemeral CLI tools \(Cursor, Claude Code\) but architecturally wrong for server-side agents that must survive restarts.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T06:44:44.177434+00:00— report_created — created