Repository navigation
v0.11.2
macOS: screenshots no longer time out when two servers share a binary
Every screenshot, zoom and OCR pass could fail with ScreenCaptureKit did not answer within 20 s while Screen Recording was granted and screencapture from a shell still worked. It began as soon as a second server started from the same binary (the npx cache, an app's bundled copy, or a local build) took a screenshot. It lasted until one of the two exited.
The cause: replayd, the macOS daemon behind ScreenCaptureKit, keeps one connection per client and identifies a bare binary by its path. Two servers from one file kept evicting each other's connection, thousands of times a second, so neither ever got an answer, and replayd ran at over 100% CPU.
- Each server now runs from its own copy, so replayd sees one client per process. On APFS the copy is a clone and takes no disk space. It is removed when the server exits. Set
MUNIM_COMPUTER_USE_SHARED_EXECUTABLE=1to turn this off. - If ScreenCaptureKit stops answering for any other reason, captures switch to
/usr/sbin/screencapturefor a minute, then try ScreenCaptureKit again. A hang now costs one slow capture instead of failing every capture after it.
A server that is already stuck keeps failing until it restarts. Reconnect the MCP server (or restart the session) after updating.