›_ Claude Code Issues
Claude Code Issues / #39270

[BUG] Claude Code process exited with code 1

Open 💬 26 comments Opened Mar 26, 2026 by data357

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

Environment

  • macOS: Darwin 25.3.0 (Apple Silicon / arm64)
  • Claude Desktop: v1.1.8629
  • Bundled Claude Code binary: v2.1.78
  • CLI Claude Code: v2.1.84 (works fine)
  • Node: v24.14.0
  • Git: 2.50.1 (Apple Git-155)

### Description

Every message sent in Claude Desktop fails silently — the message appears briefly, shows "working", then disappears
with no response. This started on March 25, 2026, coinciding with the major Anthropic outage.

The local Claude Code process (v2.1.78) that the Desktop app spawns crashes immediately with exit code 1 on every
message. The same binary works perfectly when invoked directly from the terminal with identical flags.

### Logs

main.log — repeats on every message attempt:
``
Session local_XXXX query error: Claude Code process exited with code 1
[CCD CycleHealth] unhealthy cycle for local_XXXX (0s, hadFirstResponse=false, reason=no_response)
``

unknown-window.log — CSP violation blocking API:
``
Connecting to 'https://a-api.anthropic.com/v1/m' violates the following Content Security Policy directive:
"connect-src 'self' https://api.segment.io https://*.segment.io https://*.segment.com https://a-cdn.anthropic.com
https://*.google.com https://*.facebook.com https://*.facebook.net https://*.doubleclick.net"
Fetch API cannot load https://a-api.anthropic.com/v1/m. Refused to connect because it violates the document's Content
Security Policy.
``

### Steps to Reproduce

  1. Open Claude Desktop v1.1.8629
  2. Select any project folder
  3. Send any message
  4. Message shows "working" briefly, then disappears — no response

### What I've tried (none resolved the issue)

  • Full quit and restart of Claude Desktop
  • Clearing all caches (Cache, GPUCache, Code Cache, etc.)
  • Deleting and re-downloading Claude Desktop (same version 1.1.8629 served)
  • Clearing all session data and worktrees
  • Removing and letting skills plugin re-download
  • Full reset of ~/Library/Application Support/Claude/ (keeping only login credentials)
  • Re-logging in
  • Accepting Xcode CLI Tools license (sudo xcode-select --reset)
  • Enabling Terminal in Developer Tools (Privacy & Security)
  • Installing Rosetta 2

### Key observation

The bundled binary works when run directly from the terminal:
``bash
"/Users/.../Claude/claude-code/2.1.78/claude.app/Contents/MacOS/claude" \
--print "hi" --model "claude-opus-4-6[1m]" --permission-mode "acceptEdits" \
--plugin-dir "/Users/.../skills-plugin/..."
# Output: "Hi! How can I help you today?" ✅
``

But when the Desktop app spawns it as a child process, it exits with code 1 immediately. This suggests the issue is in
how the Desktop app passes environment/auth/SDK parameters to the Claude Code process — not the binary itself.

### Additional context

  • The CSP policy in unknown-window.log does not include a-api.anthropic.com, which may be blocking telemetry

or API calls from the webview.

  • The issue began during the March 25 Anthropic outage but persists after services recovered.
  • Claude Code CLI (claude in terminal) and claude.ai in the browser both work fine — only the Desktop app's local

process integration is broken.

What Should Happen?

### Expected behavior
Messages should be sent and receive a response in Claude Desktop.

### Actual behavior
Claude Code process crashes instantly (exit code 1), message disappears, no response is shown.

Error Messages/Logs

Steps to Reproduce

Open Claude Desktop v1.1.8629 on macOS (Apple Silicon / arm64, Darwin 25.3.0)

  1. Select any project folder as the working directory
  2. Send any message (e.g., "hello")
  3. Observe: message shows "working" briefly, then disappears with no response

Note: The bundled Claude Code binary (v2.1.78) works when invoked directly from terminal:

"/Users/.../Claude/claude-code/2.1.78/claude.app/Contents/MacOS/claude" \
--print "hi" --model "claude-opus-4-6[1m]" --permission-mode "acceptEdits" \
--plugin-dir "/Users/.../skills-plugin/..."
# Output: "Hi! How can I help you today?" ✅

The crash only occurs when the Desktop app spawns the process internally.

Expected behavior:
Message is sent and a response is displayed in the conversation.

Actual behavior:
The local Claude Code process crashes instantly with exit code 1. The message disappears and no response is shown.

main.log shows on every attempt:
Session local_XXXX query error: Claude Code process exited with code 1
[CCD CycleHealth] unhealthy cycle for local_XXXX (0s, hadFirstResponse=false, reason=no_response)

unknown-window.log shows a CSP violation blocking the API:
Connecting to 'https://a-api.anthropic.com/v1/m' violates the following Content Security Policy directive
Fetch API cannot load https://a-api.anthropic.com/v1/m. Refused to connect.

The CSP connect-src directive does not include a-api.anthropic.com.

Additional context:
Environment:

  • macOS Darwin 25.3.0 (Apple Silicon / arm64)
  • Claude Desktop v1.1.8629
  • Bundled Claude Code binary: v2.1.78
  • CLI Claude Code: v2.1.84 (works fine)
  • Node: v24.14.0 / Git: 2.50.1 (Apple Git-155)

Timeline: Issue started March 25, 2026 during the major Anthropic outage (4,000+ Downdetector reports). Persists after
services recovered.

Troubleshooting attempted (none resolved):

  • Full quit/restart of Claude Desktop
  • Clearing all caches, sessions, worktrees, and skills plugins
  • Full reset of ~/Library/Application Support/Claude/ (preserving login)
  • Re-downloading Claude Desktop (same v1.1.8629 served from claude.ai/download)
  • Re-authenticating after full reset
  • sudo xcode-select --reset
  • Enabling Terminal in macOS Developer Tools
  • Installing Rosetta 2

Claude Code CLI and claude.ai in browser both work — only the Desktop app's local process integration is broken.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

v1.1.8629

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_

View original on GitHub ↗

26 Comments

github-actions[bot] · 3 months ago

Found 3 possible duplicate issues:

  1. https://github.com/anthropics/claude-code/issues/38850
  2. https://github.com/anthropics/claude-code/issues/37822
  3. https://github.com/anthropics/claude-code/issues/36419

This issue will be automatically closed as a duplicate in 3 days.

  • If your issue is a duplicate, please close it and 👍 the existing issue instead
  • To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

marketingmei · 3 months ago

Same issue, happening inside Claude.ai artifacts not just Claude Code desktop. AI-powered artifacts are completely broken since March 25

marketingmei · 3 months ago

Still broken as of March 31, 2026 – 6 days since the regression started. This is affecting AI-powered artifacts on Claude.ai, not just Claude Code desktop. The fetch to api.anthropic.com hangs indefinitely with a 30-second timeout, no error returned. This is blocking a core advertised feature.

alexfurm7 · 3 months ago

Same issue. In Claude Cowork. On an existing thread.

EmilioMoreno · 3 months ago

Same here. Exactly the same

dannytrann · 3 months ago

I'm facing the same issue today

bbelanger · 3 months ago

Ditto: Claude Cowork on first message of new chat...
Claude Code process exited with code 1

Arthurburlamaqui · 3 months ago

Same here but in claude cowork

jboling-fundeasy · 3 months ago

Same issue in Claude Code. I've tried a bunch of internal fixes; nothing works.

onbayev · 3 months ago

Same issue. Apple M1 Pro, MacOS Tahoe Version 26.4.1 (25E253)

frankfium · 3 months ago

Just updated, cowork is having this issue now

ManuelaGar · 3 months ago

Me too!

honghuy1711-a11y · 3 months ago

same here, tried several techniques that I could found but nothing work

stlblufan · 3 months ago

same issue, cowork

efraintorlo · 2 months ago

I had this same issue, this is the fix that worked in my mac https://github.com/anthropics/claude-code/issues/49391#issuecomment-4285254876

beebop1032 · 2 months ago

Kill Claude Desktop if it is still running via Terminal (or close Claude)

pkill -9 -f "Claude"

Reset macOS privacy permissions related to file access

tccutil reset SystemPolicyAllFiles

tccutil reset SystemPolicyDocumentsFolder

tccutil reset SystemPolicyDownloadsFolder

tccutil reset SystemPolicyDesktopFolder

Reset all permissions specifically for Claude Desktop

tccutil reset All com.anthropic.claudefordesktop

Accept all permissions

banosdavidhamelyn · 2 months ago

Same issue here, still happening as of April 24, 2026.

  • OS: macOS [tu versión, p.ej. Sonoma 14.5]
  • Claude Desktop version: [mira en menú Claude → About]
  • Model: Opus 4.7 with 1M context
  • Behavior: every new session in any project crashes immediately

with "Claude Code process exited with code 1". Happens with both
Opus 4.7 and Opus 4.6. CLI works fine, only the desktop app fails.

  • Repro: open any project, send any message ("hello" included),

process exits instantly.

navedsiddiqi · 2 months ago

Have exactly the same issue on Claude Code Desktop on macOS. Terminal CLI and the VS Code plugin both work fine — only the Desktop app spawns a child process that exits with code 1 immediately. Logs show the same CSP violation against a-api.anthropic.com.

elamje · 2 months ago

Same issue here. I've tried all reset/redownload/delete tricks. In Code on Claude Desktop it will run on a legacy code chat, but any new session immediately crashes.

wjm0913 · 2 months ago

no such column: paths

fate8383 · 1 month ago

the same here

ivan-zakharchuk · 1 month ago

same shit, how claude can't fix this, I'm paying PRO and no support whatsoever

prasathmani · 1 month ago

same here

AlonIsrael86 · 1 month ago

Also hitting this on Windows — Claude Desktop build 1.12603.1, bundled Claude Code agent 2.1.170. Every Code session dies the instant a message is sent (Claude Code process exited with code 4294967295, i.e. -1); the session "warms" fine, then the query process exits ~3s in. Chat/Cowork work; only Code is affected.

I ran a fairly exhaustive isolation, and it strongly confirms the "Desktop spawns the child differently than the CLI" theory — the fault is in the app's spawn, not the agent/config/env/auth:

  • Captured the app's exact spawn command line (elevated WMI) and the spawned agent's full environment (PEB read of the live child process). Replaying that exact command + exact environment + --resume <session> + the inline --mcp-config + all --plugin-dirs from a normal shell → works, exit 0.
  • The bundled claude.exe (2.1.170) also runs a real query fine standalone, including the --input-format stream-json stdin protocol.
  • So the binary, command line, environment, OAuth token, MCP servers, plugins, and even the resumed session are all fine. Only the child process the Desktop app itself spawns dies before the first response.

Ruled out (no effect): full quit/relaunch, Windows reboot, full uninstall + reinstall (the "latest" installer resolves to the same build, so reinstalling can't bring a fix), restarting Explorer to refresh the env, claude doctor (all pass).

Note for anyone arriving from the Windows duplicate (#52766): setting CLAUDE_CODE_GIT_BASH_PATH did not fix it here — it was unset, I set it and verified via PEB read that the spawned child inherited it, and it still crashes. No SessionStart/UserPromptSubmit hooks either (both empty).

Confirmed workaround: start sessions from the terminal CLI (claude) — those work fine and even show up in the Desktop session list. Only the in-app "New session" path is broken.

Happy to share the captured (secret-scrubbed) command line and environment diff if it would help narrow down what Desktop changes about the spawn.

AlonIsrael86 · 13 days ago

Anything new on this?

AlonIsrael86 · 6 days ago

Windows update with additional isolation and a supported workaround, tested July 10, 2026.

Current versions:

  • Claude Desktop: 1.20186.0.0
  • Bundled Claude Code: 2.1.205
  • Terminal Claude Code: 2.1.206
  • Failure remains: Claude Code process exited with code 4294967295

The Desktop-native spawn failure still reproduces after all of the following:

  • Updating Desktop and letting it redownload its bundled Claude Code binary
  • Full Windows reboot
  • Full Desktop logout/login with fresh OAuth authorization
  • Clean temporary working directory
  • Fixing and validating the user settings JSON encoding
  • Disabling Remote Control at startup
  • Directly testing the bundled binary with both an empty MCP configuration and the active MCP configuration
  • Isolating the local plugin registry

New evidence from restoring pre-incident session metadata:

  • Desktop can load and display historical local Code transcripts and their project/sidebar assignments correctly.
  • Sending any message from one of those restored local cards, including /compact, immediately produces Session was interrupted and the same exit code 4294967295.
  • Therefore the transcript, sidebar state, account, model entitlement, and stored session data are healthy. The failure occurs when Desktop tries to start or resume the local Claude Code child process.
  • The same underlying conversation can be resumed successfully from the terminal CLI with its history intact.

Verified workaround using Anthropic's documented Remote Control server mode:

claude remote-control --spawn same-dir --capacity 32

After selecting that Remote Control environment in Desktop, newly created sessions work end to end. A current Haiku 4.5 session successfully responded, used the local Read tool, and read workspace files through Desktop. The hub also survives a Desktop close/relaunch because it runs as an independent local process.

Practical boundary:

  • New sessions created through the Remote Control server are usable.
  • Old Desktop-local cards are readable archives but cannot be made writable in place while the native spawn bug remains.
  • Important old conversations can be resumed or forked from the CLI and exposed as new Remote Control cards.
  • Multiple same-dir sessions are not isolated from simultaneous edits, so users still need to avoid concurrent writes to the same project or use project-specific worktree servers.

This issue is currently labeled only platform:macos, but the reproducible Windows variant remains active. A platform:windows label would help keep the two manifestations visible.