All posts
Your AI can now run commands on your servers — safely
AI coding tools are great at telling you what to run. Sshly now lets them actually run it — on your real servers — while keeping you firmly in control.
The idea
Sshly already holds your connections and secrets (safely, in the Keychain). So instead of handing an AI your credentials, Sshly exposes a small MCP bridge. The AI asks for a server by name, Sshly makes the connection, runs the command, and returns only the output.
The AI never sees a password or a private key. It sees a server name and command output — nothing more.
You stay in control
- Per-server opt-in — nothing is reachable until you switch it on.
- Approve each command — Sshly asks before anything runs (or trust a server for 15 minutes).
- A hard denylist — catastrophic commands like
rm -rf /are always blocked. - Full audit log — every command the AI runs is recorded.
Setting it up takes one command in Claude Code. Open Settings → AI Access in Sshly to get started.
Ready to try the friendly SSH client?
Download Sshly