Using Hermes agent to manage several home servers safely
is running in a on a desktop, and the goal is to let it inspect and manage several Linux . The main concern is avoiding full , because that would give the agent broad power to run commands on the servers.
One option is to run an MCP server on each Linux server, exposing only the read and write actions that feel acceptable. Another option is to run a separate on each server, with limited permissions or local MCP servers, then let the talk to those remote agents through a shared chat channel, MCP, or a similar link.
The desired workflow is asking the local agent to investigate a memory usage alert on a specific server, find the likely cause, and suggest ways to fix it. Other useful abilities would include updating and s on the servers.
Key points
- The setup runs on a desktop and aims to manage several Linux .
- Full is considered too risky.
- One approach is to run an MCP server on each server with only approved actions exposed.
- Another approach is to run separate instances on each server and let them communicate with the local agent.
- The target use case is investigating a memory usage alert and suggesting fixes, plus updating or configs.