Most cloud dashboards are built for one-person-one-browser. If you manage more than a handful of servers, you end up in the terminal anyway — and once you're there, the experience suddenly drops back to 2010: tabs of SSH, ad-hoc scp, browser windows for CloudTrail, a separate CLI for each provider.

Servonaut is a TUI that stitches that back together, and then goes one step further: it exposes the same operations to AI agents like Claude Code through a Model Context Protocol server, so you can say "grep the last hour of nginx errors on prod-web-2" in natural language and get a real answer, not a hallucinated one.

What's in the box today

  • An interactive TUI (Textual-based) for listing, searching, and SSHing into instances across AWS and custom/OVH hosts
  • CloudTrail scanning and CloudWatch log analysis, with AI-assisted summaries via OpenAI / Anthropic / Ollama
  • File transfer via SCP built into the same UI
  • A local MCP server your on-machine agent (Claude Desktop, etc.) can talk to over stdio
  • A hosted MCP server at mcp.servonaut.dev that AI agents can reach via SSE, which routes tool calls back to your own CLI over a secure relay — so the agent runs commands using your SSH keys, not shared infrastructure

Why a TUI, not a web app

Because server management should not require a browser, cookies, a logout flow, and a fresh TLS handshake every few minutes. The TUI starts in 200ms, keeps SSH sessions warm, and does not forget where you were when you come back. The web site is for signup, billing, and documentation — and for AI agents to authenticate against — not for doing the actual work.

Free and open source

The CLI is MIT-licensed and will stay that way. The paid tiers add cloud config sync, AI quotas, and the hosted MCP server for teams that want AI agents to operate fleets without exposing long-lived credentials to third parties.

Grab it with pipx install servonaut, or head to the installation docs for the full options.