fix: increase MCP idle timeout to 10 minutes (#148)

[justrach]

Summary

Fixes #148 — MCP server disconnects during debugging sessions + orphaned processes linger.

Two fixes:

1. Idle timeout: 2min → 10min
Agents thinking for >2min during debugging would lose the MCP connection. 10 minutes is enough for any debugging session.

2. stdin poll for dead client detection
Watchdog now polls stdin every 10 seconds for POLLHUP. When the client pipe closes (crash, kill, disconnect), server exits within 10 seconds — no more dangling processes.

Behavior matrix:

Scenario	What happens	Exit time
Client disconnects cleanly	stdin EOF	Immediate
Client crashes/killed	POLLHUP detected	≤10 seconds
Client connected but idle	10min timeout	10 minutes
Orphaned (no client ever)	10min timeout	10 minutes

Why this is safe:

• stdin EOF still exits immediately via the main MCP read loop
• POLLHUP catches cases where stdin stays "open" but the client is gone
• 10min is generous enough for debugging but still cleans up orphans

1 file changed, ~20 lines.

[github-actions]

Benchmark Regression Report

Threshold: 10.00%

Tool	Base (ns)	Head (ns)	Delta	Status
codedb_bundle	621987	626185	+0.67%	OK
codedb_changes	108751	117851	+8.37%	OK
codedb_deps	30720	32003	+4.18%	OK
codedb_edit	17535	17943	+2.33%	OK
codedb_hot	161938	160968	-0.60%	OK
codedb_outline	401253	407089	+1.45%	OK
codedb_read	170009	166593	-2.01%	OK
codedb_search	255751	262893	+2.79%	OK
codedb_snapshot	3462658	3534808	+2.08%	OK
codedb_status	28285	29226	+3.33%	OK
codedb_symbol	62089	61366	-1.16%	OK
codedb_tree	80249	80937	+0.86%	OK
codedb_word	92130	92678	+0.59%	OK

[github-actions]

Benchmark Regression Report

Threshold: 10.00%

Tool	Base (ns)	Head (ns)	Delta	Status
codedb_bundle	645800	615901	-4.63%	OK
codedb_changes	111295	107266	-3.62%	OK
codedb_deps	34465	29855	-13.38%	OK
codedb_edit	17476	17586	+0.63%	OK
codedb_hot	160919	162053	+0.70%	OK
codedb_outline	383892	383679	-0.06%	OK
codedb_read	165333	171138	+3.51%	OK
codedb_search	256911	252397	-1.76%	OK
codedb_snapshot	3501637	3469508	-0.92%	OK
codedb_status	29236	29389	+0.52%	OK
codedb_symbol	62081	59374	-4.36%	OK
codedb_tree	81020	85217	+5.18%	OK
codedb_word	92507	90240	-2.45%	OK

[github-actions]

Benchmark Regression Report

Threshold: 10.00%

Tool	Base (ns)	Head (ns)	Delta	Status
codedb_bundle	614701	612400	-0.37%	OK
codedb_changes	109266	108125	-1.04%	OK
codedb_deps	29699	31582	+6.34%	OK
codedb_edit	17162	15987	-6.85%	OK
codedb_hot	158480	159936	+0.92%	OK
codedb_outline	403952	384902	-4.72%	OK
codedb_read	164353	171103	+4.11%	OK
codedb_search	252426	257065	+1.84%	OK
codedb_snapshot	3464697	3491893	+0.78%	OK
codedb_status	30412	27917	-8.20%	OK
codedb_symbol	62942	61583	-2.16%	OK
codedb_tree	80514	85588	+6.30%	OK
codedb_word	90855	89491	-1.50%	OK

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 252a9f1fed

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Add a real timeout when waiting for MCP child exit

This test claims to "wait up to 15 seconds," but child.wait() is unbounded and can block forever. If codedb mcp fails to exit after stdin closes (the exact regression this test is meant to catch), the test will hang the suite/CI instead of failing fast. Please use a deadline-based wait loop (e.g., tryWait + sleep, then kill/fail at 15s) so timeouts are enforced.

Useful? React with 👍 / 👎.