Failed to Start Workspace: Claude Cowork Cannot Connect

The spinner runs. The workspace does not start. You get a message along the lines of “failed to start workspace” or “workspace still starting,” and clicking Retry produces the same result. Meanwhile you have work to do.

This one is annoying because the error message does not tell you whether the problem is your network, your machine, the app, or Anthropic’s servers. It could be any of them. Here is how to figure out which one and fix it.

What “The Workspace” Actually Is

Before troubleshooting, it helps to know what you are waiting on. When Claude Cowork runs tools (reading files, running code, browsing, using MCP servers), it does not run them directly on your operating system. It spins up a sandboxed Linux environment first. Think of it as a locked room where tools can run safely without being able to reach things they should not touch.

On macOS, this sandbox is built on Apple’s Virtualization framework. On Windows, it uses a different virtualization layer that is newer and still being refined. Either way, the sandbox has to boot before Cowork can do anything useful, and if that boot fails or the connection between the app and the sandbox drops, you get the workspace error.

The failure can happen at three distinct points: the sandbox itself fails to start, the sandbox starts but the app cannot reach it over the local network socket, or the app starts fine but cannot reach the Claude API over the internet. All three produce similar error messages.

Causes and Fixes

1. The Workspace Is Still Booting

This is the most common cause and the least alarming.

1. Click Retry in the error dialog.
2. If it fails again, wait 30 seconds and click Retry again.
3. Repeat up to four or five times. On a slow machine or after an update that requires re-downloading the sandbox image, boot can take over a minute.
4. If it eventually connects, you are done. The delay will usually be shorter on subsequent launches.

2. No Internet Connection or Intermittent Dropout

Cowork needs internet access to reach the Claude API. The workspace sandbox itself is local, but the AI model is not.

1. Open a browser and load any page to confirm you have a working connection.
2. If you are on Wi-Fi, try moving closer to your router or switching to a wired connection.
3. If you are on a mobile hotspot, try switching to a more stable connection. Intermittent packet loss is enough to break the API handshake.
4. Retry the workspace after confirming connectivity.

3. VPN or Corporate Proxy Blocking the Local Socket (Especially on Windows)

This is the most common cause on Windows in managed work environments, and it trips people up because the VPN appears to be working fine for everything else.

Cowork’s sandbox communicates with the desktop app over a local network socket (localhost). Some VPN clients route all traffic, including localhost traffic, through the tunnel. When that happens, the connection between the app and the sandbox either fails outright or adds enough latency to trigger a timeout.

1. Temporarily disconnect your VPN and retry the workspace.
2. If the workspace starts successfully without the VPN, the VPN is the cause.
3. The permanent fix depends on your VPN software. Look for a “split tunnel” or “exclude local traffic” setting that lets localhost connections bypass the tunnel. Your IT department may need to configure this for managed devices.
4. If you cannot modify VPN settings, try switching to a personal device or a different network to confirm, then report the issue to your IT team.

Corporate HTTP proxies can cause similar problems. If your machine is configured to route traffic through a proxy, check whether the proxy settings exclude localhost or 127.0.0.1. On Windows, this is in Settings, then Network and Internet, then Proxy.

4. Firewall Blocking the Workspace

Security software (Windows Defender Firewall, third-party endpoint protection, Little Snitch on macOS) can block the connections Cowork needs.

1. On Windows, open Windows Defender Firewall and look for any rules blocking Claude. If you see one, temporarily disable it and test.
2. On macOS, if you have Little Snitch or a similar app, check for any Claude-related deny rules.
3. If disabling the firewall rule fixes the issue, create a targeted allow rule for the Claude desktop app rather than leaving the firewall off.

5. Disk Space Too Low

The workspace sandbox requires several gigabytes of free disk space. If your disk is nearly full, the sandbox image cannot be extracted and the workspace fails silently.

1. On macOS, open About This Mac and check available storage.
2. On Windows, open File Explorer and check the C: drive.
3. If you have less than 5 GB free, clear space before retrying. Large files in Downloads and Trash/Recycle Bin are usually the fastest wins.
4. Restart the app after freeing space.

6. The App Needs an Update

Workspace startup failures are sometimes fixed in a new app version. Older builds can have bugs in the sandbox initialization sequence that were patched later.

1. Open the Claude desktop app menu and look for “Check for Updates” or “About.”
2. Install any available update.
3. Relaunch and retry.

Anthropic publishes release notes at https://www.anthropic.com. If there is a known workspace startup bug in your version, the fix will typically be in the most recent release.

7. Anthropic’s Servers Are Down

If everything on your end looks fine, check whether the problem is upstream.

1. Visit https://status.anthropic.com for the live status page.
2. If there is an active incident affecting the Claude API, the workspace will not connect regardless of what you do locally. Wait for Anthropic to resolve the incident and try again.
3. You can also check https://support.claude.com for any announcements about known outages.

8. Full Restart

If none of the above resolves it, the classic full-restart sequence clears a surprising number of edge cases.

1. Quit the Claude desktop app completely (Cmd+Q on macOS, right-click the tray icon and Quit on Windows).
2. Wait 10 seconds.
3. Relaunch the app.
4. If still failing, restart your machine and try again after reboot.

On Windows especially, network stack state can persist across app restarts in ways that only a full reboot clears.

Platform Notes

macOS: Workspace failures here are usually either network-related or disk-space-related. The Apple Virtualization framework is mature and rarely the cause. If you see repeated failures on a healthy machine with good connectivity, check for a firewall or Little Snitch rule.

Windows: VPN conflicts are common on Windows, and the local socket setup is sensitive to network configuration. If you are on a managed work machine, your IT team’s security tooling is the first place to look after ruling out internet connectivity.

When to Contact Support

If you have worked through all of these steps and the workspace still does not start, file a support request at https://support.claude.com. Include your operating system and version, whether you are on a managed or personal device, whether you have a VPN or proxy, and the exact error text you see. Screenshots help. The support team can check server-side logs that are not visible from the app.