Cowork: "Failed to create bridge sockets after 5 attempts" on every new task (Claude Desktop 1.3036.0, Claude Code 2.1.111, macOS 26.3.1 ARM)
Summary
Every new cowork task fails within ~1 second of spawn with:
Error: Failed to create bridge sockets after 5 attempts
This looks related to #19503 (resume-only bug) and #23746 (1.1.2156 regression, closed as duplicate), but my repro differs: new tasks, not resumes, and allowedDomains=23 — so neither root cause there matches. Still happening on the latest Claude Desktop.
Environment
- Claude Desktop: 1.3036.0
- Claude Code (in-VM SDK): 2.1.111 (sha256 prefix
99376866bf7ec367) - macOS: 26.3.1 (ARM64 / Apple Silicon)
- VM bundle version:
5680b11bcdab215cccf07e0c0bd1bd9213b0c25d
Repro
- Open Claude Desktop.
- Start a brand-new cowork task (no resume).
- Task exits ~1s later with the bridge-sockets error.
Reproduces 100% of the time across 5+ fresh VM boots.
Log evidence
From ~/Library/Logs/Claude/cowork_vm_node.log:
[Spawn:config] Creating spawn function for process=compassionate-wonderful-wozniak,
isResume=false, mounts=7 (outputs, .claude, .claude/skills, .cowork-lib,
.cowork-perm-req, .cowork-perm-resp, uploads), allowedDomains=23
[Spawn:create] ... cmd=/usr/local/bin/claude args=... --permission-prompt-tool stdio
--allow-dangerously-skip-permissions ... cwd=/sessions/compassionate-wonderful-wozniak
[Spawn:vm] OAuth token approved with MITM proxy
[Spawn:vm] Spawn succeeded in 45ms
[vm-stderr ...] Error: Failed to create bridge sockets after 5 attempts
[Process:...] Exited, code=1, signal=null, duration=948ms, oom=false
VM side (coworkd.log) shows clean boot: gvisor network up, static IP assigned, MITM proxy started on /var/run/mitm-proxy.sock, OAuth token approved. Failure is entirely inside the spawned claude process.
Root of the error is Claude Code's own sandbox bridge setup (u57() in the bundled JS) — it spawns two socat processes for HTTP/SOCKS unix-to-TCP bridges at /tmp/claude-http-*.sock and /tmp/claude-socks-*.sock and polls 5× for them to appear. They never do.
What I've tried (no change)
- Full quit + relaunch of Claude Desktop
- Multiple fresh VM boots
- Removing
/tmp/claude-*on host - Confirmed new tasks only, no resume
- Deleted \
~/Library/Application Support/Claude/claude-code-vm/\(forces SDK re-download) - Deleted \
~/Library/Application Support/Claude/vm_bundles/\(forces full rootfs re-download)
Neither reset changed anything — fails identically after re-download.
Happy to grab further diagnostics if there's something specific you'd like me to run inside the VM.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗