[poolski]

Description

When a core exits without cleaning up (crash, SIGKILL, the app being replaced while it runs), it leaves ~/.mcpproxy/mcpproxy.sock behind. On the next start the tray probes that socket, gets a refusal, and treats it as a core that may be there: it waits out unresponsiveCoreTimeout (60 s) and only then checks the data directory lock (DataDirectoryLock.probe), finds it free, and launches a core. Every start after an unclean exit takes a minute before the core comes up.

start() now checks the lock up front. If every probe has refused (socketEvidence.onlyEverRefused) and nobody holds config.db, it launches immediately through the existing spawn path (confirmNothingIsListening, then launchWithRetries). The tray still never unlinks the socket; the core's cleanupStaleSocket removes it.

The 60 s wait stays for every other case:

• the lock is held: bbolt takes it before a core binds its listener, so a live core, including a saturated one (GH tray: a live core with a saturated listen backlog can be launched over, producing doomed spawns and a misleading error #933), is never launched over;
• the lock state is .unknown, for example a socket with no database beside it;
• spawning is not allowed (maySpawn false).

Testing

• SaturatedCoreEscalationTests: a refusing socket beside an unheld config.db launches within 2 s with a 30 s deadline (fails without the change), and a refusing socket with no database still waits. The existing GH tray: a live core with a saturated listen backlog can be launched over, producing doomed spawns and a misleading error #933 cases are unchanged.
• Full tray suite: swift test in native/macos/MCPProxy

[Dumbris]

Thanks — the lock-first shortcut is exactly right and keeps the #933 guarantees. Review found only test hygiene, which I pushed on top: the two CoreLivenessTests that leave a refusing socket in bare /tmp now use a private directory (the derived lock path was /tmp/config.db, so a stray unheld file there would have made the shortcut fire), a maySpawn: false case for the new condition, and two doc comments updated for the new path. Full swift test is 1755/0. Auto-merge is armed.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[Dumbris]

Merged. Thank you, @poolski! Every tray start after a crash or SIGKILL used to sit a full minute on the dead core's socket; checking the data-directory lock up front turns that into an immediate launch while keeping the saturated-core guarantee from #933 intact.

Contributions like this are what keep MCPProxy moving, and they matter a lot to the project.

One small ask, only if you feel like it: a short public note about your change (a tweet, a LinkedIn post, a line in a community you are part of) helps more people discover MCPProxy. No pressure at all, and I would be glad to repost it.