Bug

MCP server exits with code=0 shortly after startup or after first tool call, causing MCP error -32000: Connection closed on every subsequent call.

Root cause

The fix from #236 (commit bfbf654) — which removed process.stdin.on("end", ...) from the lifecycle guard — was reverted in commit 2e44cf6. The revert brought the stdin listener back.

PR #311 (v1.0.111) added grandparent ppid detection but explicitly noted "stdin intentionally not touched." However, server.bundle.mjs in v1.0.111 still contains the reverted stdin listener:

process.stdin.isTTY||process.stdin.on("end",a)

This means the original #236 bug is back: transient stdin pipe events trigger gracefulShutdown() → process.exit(0), and the MCP client logs -32000: Connection closed.

Evidence from spy log

=== START PID=306114 ===
initialize → OK response
notifications/initialized → received
EXIT code=0   ← 120ms after init, before tools/list even arrives

=== START PID=306161 (after /mcp reconnect) ===
initialize → OK
tools/list → OK  
ctx_stats call → response sent
EXIT code=0   ← ~10s after last response

Pattern: server starts, processes 0-1 requests, then exits cleanly. Every /mcp reconnect starts a new process that dies the same way.

Timeline

Fix

Remove the stdin end listener from the lifecycle guard, exactly as #236 originally did. ppid polling (30s) + SIGTERM/SIGINT/SIGHUP signals are sufficient for orphan detection.

In src/lifecycle.ts, remove:

process.stdin.isTTY || process.stdin.on("end", a)

And rebuild server.bundle.mjs.

Workaround

Patch server.bundle.mjs directly:

- process.stdin.isTTY||process.stdin.on("end",a)
+ /* PATCHED */ void 0

Environment

• context-mode v1.0.111
• Claude Code v2.1.92
• Linux 6.8.0-110-generic
• Node v22.x