Your homelab, legible to AI agents — a read-only MCP server that rides along with your monitor

Description and use

A read‑only MCP server exposes homelab monitoring data to AI agents, enabling accurate queries of GPU usage, energy cost, and disk usage without manual data transfer.

Problem

AI agents could not access detailed homelab monitoring data, forcing manual data transfer and inaccurate inference.

Outcome

Agents can now retrieve accurate, real‑time homelab metrics without manual intervention, while remaining read‑only for security.

Workflow

• 1. Run monitoring container that collects data into SQLite.
• 2. Enable MCP server (enable_mcp=0 turns it off).
• 3. Start MCP server on port 9810 inside the container.
• 4. Register MCP endpoint with an agent using claude mcp add --transport http homelab http://YOUR-HUB:9810/mcp.
• 5. Agent queries tools such as hosts, containers, services, per‑card GPU state, energy cost, etc. to obtain read‑only insights.

Evidence

i keep a monitoring container on my homelab — it reads nvidia-smi, the docker api, systemd and /proc, and turns it into one page.

Source

the agent couldn't see my homelab. it could run commands, but it couldn't see — which model is holding the 3090, why the temperature is climbing, what the month cost in energy.

Source

so i added a read-only MCP server to the same container. it comes up on port 9810, and the whole setup is one line:

Source

nineteen named tools, all read-only: hosts, containers, services, per-card gpu state and which process is on which card, per-process ram, the model servers, installed models, energy cost, benchmark runs, disk treemaps, history, alerts.

Source

no writes. both directions are question and answer. the agent can look around the lab, it can't restart anything.

Source