[FEATURE] /sandbox command support in Claude Desktop app

Open 💬 1 comment Opened Jun 25, 2026 by nate-selzer
Re-opening #61234, which was auto-closed as stale. The request below is unchanged.

Summary

The /sandbox command is available in Claude Code CLI but does not exist in Claude Desktop (with code), leaving users with no way to manage sandbox settings interactively from within the app.

Current Behavior

  • Bash commands to localhost / network endpoints are blocked by the harness-level sandbox in Claude Desktop.
  • The only bypass is dangerouslyDisableSandbox: true per call, which triggers a permission pop-up with only "Allow Once" / "Deny" — no "Always Allow" option.
  • sandbox.network.allowedDomains set in ~/.claude/settings.json is not honored by the Desktop harness, even when sandbox.enabled: false is set.
  • skipDangerousModePermissionPrompt: true in settings.json does not suppress the per-call pop-up in Desktop.

Expected Behavior

One or more of the following:

  1. Expose /sandbox in Claude Desktop so users can toggle sandbox mode and configure allowed domains interactively, matching the CLI experience.
  2. Honor sandbox.network.allowedDomains from ~/.claude/settings.json in the Desktop harness so users can whitelist localhost / 127.0.0.1 without needing per-call bypasses.
  3. Add an "Always Allow" option to the dangerouslyDisableSandbox permission pop-up so users can permanently authorize the bypass without repeated prompts.

Use Case

Developers using Claude Desktop for active coding work (e.g., testing local backends, running curl against localhost services) are forced to click "Allow Once" on every network call. This is a significant friction point for any workflow involving local servers.

Environment

  • Claude Desktop (macOS) with code/cowork feature
  • ~/.claude/settings.json with sandbox.enabled: false — works for CLI but not Desktop harness

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗