SikamikanikoBG/homelab-monitor
View on GitHub ↗Plug-and-play homelab dashboard in one container — GPU, local-AI VRAM, Docker, systemd, host health. Built-in read-only MCP server so AI agents can explore it too.
182 ★24 forksPythonUpdated 18d ago
What you need to know
Self-hosted one-page dashboard for your homelab and AI rig showing GPU truth, tokens/sec, power cost by the hour, uptime, containers and disks, with a built-in read-only MCP server
Install
Grab the docker-compose.yml and run docker compose up -d
Usage
- •Open the dashboard on port 9800
- •Connect an agent with claude mcp add --transport http homelab http://YOUR-HUB:9810/mcp
- •Add remote hosts by pasting the hub's SSH key
Key features
- ✓Built-in read-only MCP server with 19 named tools
- ✓GPU monitoring for NVIDIA, AMD and Intel across the fleet
- ✓Power-to-money costs per machine, component, process and container
- ✓Benchmark Lab measuring ollama model tokens/sec and VRAM fit
- ✓Multi-machine over SSH with no agents installed
- ✓Containers, systemd services, disk treemaps and uptime checks
- ✓Push alerts to Discord, ntfy.sh and Telegram
Best for
Home lab owners and AI rig builders wanting one dashboard plus MCP access to their whole fleet
Caveats
- ⚠Read-only by design - no write tools, cannot touch the fleet
- ⚠No login or auth on the dashboard - keep behind LAN/VPN
- ⚠Runs with broad host access including Docker socket by default
Platforms: Linux · macOS · WindowsClients: Claude · Claude Code · ChatGPT · Any MCP client
Reviewed 2026-08-11
Topics
agenticai-agentsai-infrastructureclaudedashboarddockergpugpu-monitoringhomelabllmlocal-aimcpmcp-servermodel-context-protocolmonitoringnvidiaollamaself-hostedsysadminvram
- Stars
- 182★
- Forks
- 24
- Language
- Python
- License
- MIT
- Created
- 2026-05-25
- Last push
- 2026-08-17