Subagent and MCP server processes are not reclaimed after task completion, causing high memory usage

Problem

Claude Code maintains multiple Node.js child processes (subagents, MCP servers) throughout a session's lifetime, even after the corresponding tasks have completed. These idle processes continue to consume memory unnecessarily.

Observed Behavior

• When subagents finish their tasks, the spawned Node.js processes remain alive instead of being terminated
• MCP server processes are started at session init and persist until session end, regardless of whether they are actively being used
• Over the course of a long session, memory usage grows significantly due to accumulated idle processes
• Each Node.js process carries ~30-50MB of baseline overhead (V8 heap, loaded modules, event loop), so a handful of orphaned processes can easily add hundreds of MB

Expected Behavior

• Subagent processes should be terminated promptly after their task completes
• MCP server processes should support idle timeout — spin up on demand, shut down after a configurable period of inactivity
• Overall process lifecycle should be actively managed to prevent unnecessary memory accumulation

Reproduction

1. Start a Claude Code session with multiple MCP servers configured
2. Run several tasks that spawn subagents
3. After tasks complete, check process list (tasklist | grep node on Windows, ps aux | grep node on Unix)
4. Observe that completed subagent processes and unused MCP server processes are still running

Suggestion

• Implement process cleanup for subagents upon task completion
• Add idle timeout mechanism for MCP servers (e.g., terminate after N minutes of no tool calls, restart on next invocation)
• Consider a process pool with an upper bound to cap total memory usage