MCP SSH server lets agents run commands on servers without ever holding keys
NoStrawberry1162 · reddit · 2026-08-17
A maker (disclosure: the author) shared an SSH client with a built-in MCP server that lets AI agents open sessions, run commands and move files on remote servers — without the agent ever holding a key.
Key design points:
- Agents get tools (hostslist, sshexec, SFTP, sessions) over MCP.
- A key custodian authenticates: unlock keys once, it signs for the agent; no key file to read.
- Per-host full/allowlist/blocked policies; per-key scope + expiry on the hosted endpoint.
- Every session mirrors live in a "watch grid" plus audit log and recording.
- Bundled local stdio server + hosted endpoint with short-lived certs; listed in the official registry as in.termal/termalin-web.
He asks for feedback: is per-host + per-command the right granularity, or tool-call-level policy? And for human-in-the-loop — approval-per-action, or watch-and-interrupt?
More from coding & agent
- MIT paper: clinical multi-agent committees cascade shortcuts; only referee agents catch it — MIT · 2026-08-17
- Agent Optimization Hierarchy: Architecture Over Human Review — MikeBirdTech · 2026-08-17
- GitHub Repo: Complete roadmap to become an AI Engineer in 2026 — ZabihullahAtal · 2026-08-17
- Anthropic Releases Dataset of 817 Cybersecurity Skills for AI Agents — mukul975 · 2026-08-17
- MoneyPrinterTurbo passes 105K stars: turn any keyword into HD short videos — harry0703 · 2026-08-17
- Test shows Sol and GLM-5.3 catch critical code flaws; Claude misses them — morgymcg · 2026-08-17