Block AI agents reading your SSH keys and credentials
Your agent works in one project. Muzzle stops its commands and MCP tools reading the keys, tokens and passwords the rest of your Mac keeps, and you allow the ones it needs with one click.
● Bash(cat ~/.ssh/id_ed25519 | base64)
⎿Muzzle blocked this: it reads your SSH key (~/.ssh/id_ed25519). To allow it, click Allow on this entry in Muzzle's Activity.
● Bash(security find-generic-password -s github -w)
⎿Muzzle blocked this: it reads your keychain passwords. To allow it, click Allow on this entry in Muzzle's Activity.
What your agent can't read
SSH keys
Block~/.ssh private keys; .pub, known_hosts and config stay readableCloud credentialsAWS, Google Cloud, Azure, Kubernetes
BlockTool tokensDocker, npm, PyPI, Cargo, GitHub CLI,
Block.netrc, .git-credentialsGPG keys
Block~/.gnupgBrowser passwords and cookiesChrome, Arc, Brave, Edge, Firefox, Safari cookies
BlockThe keychain
Blocksecurity find-generic-password -w, dump-keychain, export.env files from other projectsThe project you're in, and templates like
Block.env.example, stay readableCopies and uploads count. Using the key doesn't.
Your agent can't print, copy, pack, encode or upload these files, from a command, a script or an MCP tool. Logging in with a key, as in scp -i, still works.
Questions
My agent needs one of these files. Now what?
Click Allow on the blocked entry in Muzzle's Activity. That file, or that keychain item, is readable from then on.
Does it work with Codex?
Yes, with Pro. On Free it covers Claude Code.
Can I switch it off?
Yes, with the Credential files switch in Settings.
Related
Put Muzzle between your agents and the internet
Download for Mac
Free · Apple Silicon · macOS 12 or later