
Stack services
See only the apps you installed, with app-owned settings and helpers tucked away until you need them.
Agent-managed homelab control
Stackarr gives trusted agents one typed, safety-controlled interface to your self-hosted apps, backups, containers, downloads, and infrastructure.
Chat-first, never chat-only. Keep the dashboard and CLI, or grant an agent complete control when you choose.
Check the stack, fix what is safe, and ask before interrupting playback.
The first way in
Every client reaches the same shared actions and safety policy. Stackarr trims the catalog to the apps you installed, so agents see useful tools instead of a wall of irrelevant commands.
Add Stackarr as a local MCP server, including native MCP connections for Hermes and OpenClaw.
Open agent setup →Use observe, manage, admin, or unrestricted. The launch profile—not the agent—sets the boundary.
Compare MCP profiles →Ask for an install, a health check, a media request, a repair, or a full stack migration.
Connect Hermes or OpenClaw →Manual control, always available
The dashboard is an optional but complete operating surface for setup, status, activity, recovery, and direct service access. Agent and human actions share the same control plane.


See only the apps you installed, with app-owned settings and helpers tucked away until you need them.

Tune the experience to your setup with theme, refresh, service-link, and onboarding preferences that stay with the app.
Run it, then hand over the keys you choose
Download the Compose file and run the control plane on any Docker-compatible host.
curl -fsSL https://stackarr.app/docker-compose.yml -o docker-compose.yml
docker compose --profile stackarr up -d appStart with approval-backed management, use admin for setup, or deliberately grant full autonomy.
codex mcp add stackarr -- \
docker exec -i -e STACKARR_MCP_PROFILE=manage \
app /app/bin/stackarr mcp serveSee the MCP connection guides for Claude, Hermes, OpenClaw, and LM Studio.
Inspect, configure, repair, and operate the stack from Codex, Claude, Hermes, OpenClaw, LM Studio, or another MCP client.
Agents use typed Stackarr and app API actions instead of improvising shell commands against your containers.
Start read-only, allow routine management, unlock admin setup, or deliberately grant unrestricted autonomous control.
The dashboard, API, and CLI remain available whenever you want to inspect a plan, intervene, or work manually.