Claude Desktop crashes on macOS Tahoe 26.3.1 — Crashpad mach_port_request_notification failure

Status Fixed / completed
Maintainer reply None cached
Activity 5 comments · opened Mar 21, 2026 · closed Mar 28, 2026

Summary

Claude Desktop (v1.1.7714) crashes repeatedly on macOS Tahoe 26.3.1. Started the week of March 12, 2026. Same hardware/OS configuration works fine on a second machine.

Environment

  • macOS: Tahoe 26.3.1, Build 25D771280a
  • Hardware: Apple M4 Pro, 24GB RAM (Mac16,7)
  • Claude Desktop: v1.1.7714, Electron 40.4.1
  • Installed via: Homebrew Cask

Root Cause

Crashpad (chrome_crashpad_handler) fails on mach_port_request_notification with KERN_INVALID_CAPABILITY (error 20). This error fires every ~1 second and eventually kills the main Electron process.

Exact error (repeats every second):

[FATAL:third_party/crashpad/crashpad/handler/mac/exception_handler_server.cc:76]
Check failed: kr == KERN_SUCCESS. mach_port_request_notification: (os/kern) invalid capability (20)

Crash report exception:

"exception": {
  "codes": "0x0000000000000001, 0x00000001004e7380",
  "type": "EXC_BREAKPOINT",
  "signal": "SIGTRAP"
}

Behavior

  • App launches, shows login screen briefly, then crashes (15-30 seconds)
  • Crashes whether logged in or not
  • When crashpad binary is physically removed, app survives much longer (2+ minutes, completed login and Co-Work session), confirming crashpad as the cause
  • However, modifying the binary breaks code signature → "invalid installation" error, so this is not a viable user workaround

What Was Tried (all failed to resolve)

  1. Full uninstall/reinstall via brew reinstall --cask claude (multiple times)
  2. Complete deletion of ~/Library/Application Support/Claude/
  3. Deleting Cookies, Preferences, Keychain entries (Claude Safe Storage)
  4. tccutil reset All com.anthropic.claudefordesktop
  5. Removing quarantine: xattr -cr /Applications/Claude.app
  6. Launch flags: --disable-gpu, --disable-crash-reporter, --password-store=basic, --disable-breakpad
  7. Environment variables: ELECTRON_DISABLE_CRASH_REPORTER=1

Suggested Fix

Update Crashpad to handle KERN_INVALID_CAPABILITY gracefully on macOS Tahoe, or add error recovery so the FATAL assertion doesn't propagate to the main process.

Diagnostic Evidence

  • 79 crash reports in ~/Library/Logs/DiagnosticReports/chrome_crashpad_handler-*.ips
  • Full stderr capture showing the FATAL loop available on request

View original on GitHub ↗

5 Comments

bstanford29 · 6 months ago

Update: Still reproducible on v1.1.8308

Upgraded to v1.1.8308 today (2026-03-23) — same crash, same error, still unfixed.

[0323/195541.690985:FATAL:third_party/crashpad/crashpad/handler/mac/exception_handler_server.cc:76] Check failed: kr == KERN_SUCCESS. mach_port_request_notification: (os/kern) invalid capability (20)

This repeats every ~1 second and eventually kills the main process. Full fresh install (deleted app + all app data, reinstalled via brew), same result.

Also flagging: this issue was labeled invalid but it is a reproducible, hardware-confirmed crash. The mach_port_request_notification FATAL in Crashpad's exception handler server is the direct cause — not a misconfiguration or user error. Would appreciate a re-triage.

bstanford29 · 6 months ago

Wanted to add some urgency here — I'm on the Max plan ($200/month) and Claude Desktop has been completely unusable for 11 days now (since ~March 12). Co-Work and Dispatch are desktop/iOS only features, so I'm locked out of core functionality I'm paying for.

This issue was auto-labeled invalid but the crash is 100% reproducible, survives full reinstalls, and still occurs on v1.1.8308 (latest). Would really appreciate a human review and a status update on whether a fix is in progress.

bstanford29 · 6 months ago

Requesting re-triage: invalid label is incorrect

This issue was auto-labeled invalid but is a reproducible crash with a clear stack trace. Requesting label change to bug.

Evidence attached

1. macOS Crash Report (Claude-2026-03-21-155639.ips)

  • EXC_BREAKPOINT / SIGTRAP in Crashpad's mach_port_request_notification
  • CoreFoundation: *** multi-threaded process forked ***
  • libsystem_c.dylib: crashed on child side of fork pre-exec
  • App dies within 0.18 minutes of launch, every time

2. System Info

  • macOS 26.3.1 (25D771280a), Apple M4 Pro, 24GB RAM
  • Claude Desktop v1.1.7714 → v1.1.8308 (same crash on both)
  • Electron 40.4.1, Chrome 144.0.7559.173, Node 24.13.0

3. Reproduction

  • Launch Claude Desktop → crash within seconds
  • Survives: full uninstall + reinstall (brew), deleting all app data, rebooting
  • Has persisted for 12+ days across multiple app versions

Root cause hypothesis

The "crashed on child side of fork pre-exec" ASI annotation combined with the Crashpad mach_port_request_notification failure suggests Electron's Crashpad handler is forking a child process that inherits mach port state incompatible with macOS Tahoe 26.3.1's kernel. This is a known class of Chromium issues on newer macOS versions.

Impact

Max plan subscriber ($200/mo) — Claude Desktop completely unusable since ~March 12. Co-Work and Dispatch features are desktop-only, so this blocks core paid functionality.

bstanford29 · 6 months ago

Cross-referencing duplicate reports

This issue is not isolated — multiple users are reporting the same Crashpad mach_port_request_notification crash across macOS Tahoe versions:

  • #32170 — Claude Desktop on Mac crashing on opening
  • #28590 — Failure to Launch on Tahoe 26.3
  • #29625 — Crashes on launch within ~13s, macOS 26.4 beta
  • #19068 — Pointer authentication trap on macOS 26.2

All share the same root cause: Crashpad's exception_handler_server.cc:76 fails with KERN_INVALID_CAPABILITY (20) on macOS 26's updated Mach port APIs. The FATAL assertion fires every ~1 second and eventually kills the Electron process.

This issue has been open 14 days with no Anthropic response and is incorrectly labeled invalid. Requesting re-triage to bug.

github-actions[bot] · 6 months ago

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.