The Domotz plugin for Claude and ChatGPT: ready-made playbooks that let your AI assistant do real network-operations work in your Domotz account. It reads the data, explains what it means, and makes only the changes you approve. The plugin bundles the playbooks and the Domotz connector, so once it is listed in your assistant's plugin directory you add it, sign in to Domotz, and ask.
Works in Claude and ChatGPT; other MCP clients can paste the playbook. Each playbook is a single instruction file (an Agent Skill). There is no script to run: connect your assistant to the Domotz MCP server, add the playbook, and ask.
| Playbook | What you get | Changes it can make |
|---|---|---|
| Alert tune-up | Which alerts are noise, which rules need a better threshold, which are real signal, with evidence and ready-to-approve fixes (example) | Unbind rules, adjust thresholds (only after you approve, all reversible) |
| Site check-up | Is a new (or existing) site ready: what would never alert, what isn't marked important, which credentials and backups are missing, with fixes to approve | Bind "went down" rules, set importance, turn the open-port scanner on or off (only after you approve, all reversible) |
| Health report | The monthly report for a client: availability of their important devices, outages, alerts and the work you did, compared with the month before, as a client-ready page plus a technician appendix (example) | None: read-only |
More are on the way.
Playbooks need a frontier model (Claude Sonnet or Opus, or an equivalent). Small fast models follow the steps but are not reliable at the counting these audits need.
You need a Domotz account and an AI assistant that can use MCP connectors.
Claude (web and desktop) · ChatGPT · Claude Code · Other MCP clients
As a plugin: this repository is a plugin for Claude (.claude-plugin/plugin.json) and for
ChatGPT (plugin.json, Agent Plugins format) that bundles the playbooks and the Domotz
connector. In Claude Code, add this repository as a plugin marketplace and install it:
/plugin marketplace add domotz/domotz-plugin
/plugin install domotz@domotz
The playbooks then appear as /domotz:alert-tune-up, /domotz:site-check-up and
/domotz:health-report, and /mcp shows the Domotz connector, which asks you to
sign in to Domotz the first time it is used.
Claude (web and desktop), one playbook at a time:
- Download the playbook's zip, for example alert-tune-up.zip from the latest release.
- In Claude, open Customize → Skills, add a skill, and upload the zip.
- Check that the skill is switched on.
ChatGPT: download the same zip and upload it under Skills → Add → Upload from your computer.
Claude Code: copy the playbook folder into your skills directory:
git clone https://github.com/domotz/domotz-plugin
mkdir -p ~/.claude/skills
cp -r domotz-plugin/skills/alert-tune-up ~/.claude/skills/Any other assistant: paste the contents of
skills/alert-tune-up/SKILL.md as the first message of
a chat, then ask your question.
Run an alert tune-up for my client Acme for the last 14 days.
More starter prompts: prompts.md.
- The playbooks read and change data only in your own Domotz account, through the Domotz MCP
server at
https://mcp.domotz.com/mcp, signed in as you and limited to your permissions. They run inside your AI assistant, so what they read is processed by that assistant's provider under your agreement with them. - Changes happen only after you approve a numbered batch, and each one is reversible.
- They never ask for, store or send passwords or other credentials.
- The optional one-page review is an HTML page your assistant builds from
report-template.html. It loads its fonts from Google Fonts (fonts.googleapis.com) and sends nothing anywhere. - The plugin has no scripts, hooks or other servers, and nothing in this repository receives your data.
Every playbook follows the same rules (CONTRIBUTING.md):
- Read first, change later. Nothing changes in your account until you approve a numbered batch of changes.
- Only reversible changes, each with its undo shown before and after.
- Evidence, not opinions. Every recommendation cites the counts, devices and dates it is based on, and the report says exactly what data it covered.
- One client at a time. A batch never mixes organizations.
Each playbook is tested on real Domotz data before release, including applying its changes and undoing them.