AgentFleet — Human-approved fleet ops for AI agents

Let AI manage all your servers at once, with every step passing your approval

Problem

A developer building proxies had to jump between several SSH sessions and sometimes sent commands to the wrong server. Every small AI-assisted change required re-explaining each machine's setup, and existing MCP servers are mostly single-host and don't span servers. Multi-machine ops lacks a layer that gives the agent a fleet-wide view without letting it run wild.

Solution

An MCP server that treats the server fleet as one unit: users keep a shared list of hosts, roles and notes, and the agent plans against that context. Any step involving writes or risky commands must be approved by a human before execution, with a full audit trail of which host, which command, and who approved it.

Users

Indie developers and homelab operators running several VPS instances or self-hosted services.

MVP

Import SSH config into a host inventory; expose an MCP server so agents can query and plan; force write actions into approval cards; execute locally after approval; exportable action logs; no enterprise permission model.

Revenue

Free tier up to 5 servers; annual subscription unlocks unlimited servers, team approval flows and audit export.

Why now

Agents can now reliably plan multi-step ops tasks, but nobody wants to hand over write access unguarded — the approval layer is exactly the missing piece.