Practical Notes
- GitHub: 199 stars · 35 forks; pushed 2026-05-07 (verified via GitHub API).
- README quick start includes
npm install -g mcp-ssh-managerandclaude mcp add ... --scope user/projectoptions. - README lists 37 tools and an optional auto-approve list for reducing repeated confirmations.
Main
Treat SSH as a high-risk capability:
- Start with a read-only runbook (health checks, log tail, disk usage) and review the agent’s proposed commands.
- Use least-privilege SSH keys and separate “automation” accounts from human admin accounts.
- Keep an allowlist of hosts; avoid letting the agent accept arbitrary hostnames or copy secrets back into chat logs.
When it works well, the biggest win is consistency: every incident gets the same baseline checks before you escalate.
FAQ
Q: Can I use it without Claude Code? A: Yes—it's an MCP server; any MCP client that can spawn Node processes can use it (per README).
Q: Should I enable auto-approve? A: Only for low-risk, read-only tools; keep write actions behind manual review.
Q: How do I share configs across a team?
A: README shows using --scope project to create a project-level .mcp.json.