Claude Code security hooks

Muzzle checks each command, file write, web request and MCP call Claude Code makes before it runs, and each result after.

Set it up

  1. Download Muzzle, open it and click Get started.
  2. Enter what should never leave your Mac: your email, name and phone number. Muzzle suggests your git email and account name.
  3. Leave Claude Code switched on and click Protect. Muzzle writes its hooks to ~/.claude/settings.json and routes the commands Claude Code runs through its local proxy.
  4. Start a new Claude Code session so it loads the hooks.
  5. Muzzle scans your projects and shows anything that already leaks. For the first week it only watches, then you turn blocking on.

What's covered

Leaks in commands, file writes, web fetches and MCP calls
Free
Settings lock
Free
Credential files and package check
Free
Prompt injection alerts on results
Pro
Instruction files and risky folders at session start
Pro
Requests made by scripts Claude Code runs
Pro

Questions

How do I remove it?

Switch Claude Code off in Muzzle's Settings. Muzzle takes its hooks out of settings.json.

What if I delete the Muzzle app?

The hooks do nothing when the app is gone, so Claude Code keeps working.

Can Claude Code switch Muzzle off?

No. Muzzle blocks commands that stop it, skip its checks or route traffic around it, and the settings lock blocks edits to its settings.

Put Muzzle between your agents and the internet