Hey everyone,
I spent the last days building a small MCP → SSH relay so an LLM can safely control remote servers using a limited command set.
Here’s what the agent currently does completely autonomously:
-
Creates a temporary Hetzner server via API -
Generates its own SSH keys and stores them for the session -
Connects over SSH using my relay (no direct access for the model) -
Installs all dependencies needed for a Minecraft server -
Sets up the server and runs it inside tmux (so it stays alive) -
Adds autostart so it survives reboots -
Verifies the port is open and checks if the server responds -
Sends me a clean status summary when everything is done -
Deletes the server if something fails (or after the demo)
This all runs in a fully scripted plan the agent writes by itself —
I only give the goal (“deploy a Minecraft server on Hetzner”),
and it handles the entire workflow.
It’s still very early (pre-alpha), but the autonomy feels kinda crazy already.
I’m planning to extend this into a more general “Workbench” where agents can manage infrastructure safely.
Video demo on reddit: Reddit - The heart of the internet
Would love feedback, ideas, concerns — anything helps.