Ampio MCP

Frequently asked questions

How Ampio MCP works, what it supports and what it can do.

What it looks like in Claude Code

How it works

What is Ampio MCP?

A service that lets Claude read and program the logic of an Ampio installation. You describe what should happen in the house — "when the hall button is held, close all blinds" — and Claude turns it into rules, shows you a preview, and saves them only after you confirm.

How does Claude reach my installation?
  • A small agent (a Docker container) runs on the installation's network. It talks to the M-SERV over MQTT and keeps one outbound connection to this server — no open ports, no port forwarding.
  • Claude connects to the server as a connector (MCP over HTTPS, logged in with your account through OAuth).
  • The server holds the Ampio logic; the agent only relays the few requests the server is allowed to make (a fixed list), so it is not a general gateway into your network.
  • MQTT credentials (and Node-RED credentials, if any) stay on the agent.
What happens when Claude changes something?
  1. Claude asks for a preview: the rules before and after, warnings, and how much space the logic takes on the module.
  2. You read it and say yes — nothing is written before that.
  3. The server re-reads the module and refuses if its logic changed in the meantime, takes a backup, writes, and reads it back to confirm.

Any backup can be restored later; ask Claude to list them.

How do I start?
  1. Create an account; an administrator activates it.
  2. Add installation: run the agent shown there and enter its pairing code.
  3. Connect Claude and log in when Claude asks.
In which language does Claude name things?

In the language chosen for the installation when its agent is set up (English or Polski): rule descriptions, item, event, group and folder names and Node-RED labels that Claude creates are written in it, whatever language you talk to Claude in. The agent's own messages use it too.

Can several people or companies work on one installation?

Yes. Each account that adds the same M-SERV gets its own, independent installation (its own agent, backups and licence). Within one account, every agent on the same M-SERV joins the same installation.

What is supported

Which AI models and apps?
  • Claude — claude.ai, the Claude desktop app and Claude Code, with any Claude model the app uses (Opus, Sonnet, Haiku). Larger models handle long, multi-step changes best.
  • Applications on your own computer that connect through http://localhost (for example Claude Code) — the consent page marks them so you can tell.
  • Other services cannot connect: an authorization is only sent back to Claude or to localhost.
Which Ampio modules?
  • Current firmware (protocol 25 and newer): logic is read and written in the module's own format, exactly as the Ampio Designer does — buttons and M-DOT panels, relays and binary outputs, dimmers, LED / RGBW / CCT outputs, blinds, HVAC and climate zones, temperature, light, CO₂ and humidity sensors, flags, card readers, buzzers, LCD panels, the RTC, alarm zones and outputs, PIN keypads and Arylic audio.
  • Older modules (protocol below 25): their logic is read and written through their own path, with the same preview, backup and read-back.
  • CAN-net events and groups: as triggers and as actions of rules.
  • Rules of child devices (Satel, RS232, Zamel, Modbus) are shown; only their enabled state, description and folder can change.

The rule format follows the Ampio Designer release of November 2026.

Does it need anything else on the installation?

Only Docker on a machine in the same network as the M-SERV, for the agent. The M-SERV needs no changes.

What it can do

Logic (rules)
  • List and explain the rules of every module, in plain words.
  • Create, change, enable / disable and delete rules; several changes on one module as one save.
  • Triggers with press / hold times, counts, values and comparisons; up to three trigger blocks with AND / OR.
  • Pick the module to store a rule on when the action is a CAN-net group or event: where the group's outputs or related rules already are, in the same room, with free space, spreading logic evenly.
  • Folders like in the Designer; export and import of the Designer's rules file; copy a rule to another module.
  • Test an action once (like the Designer's test button) without saving anything.
Names, events and groups
  • Rename inputs, outputs and flags of a module.
  • Create, rename and delete CAN-net events.
  • Create CAN-net groups and change their member outputs.
Node-RED

If Node-RED runs on the M-SERV (port 1880), Claude can list and read its flows, create, change or delete one flow, and press an inject node — with the same preview, backup and read-back. Set NODERED_URL on the agent for another address, or NODERED_URL=off to turn it off.

Safety
  • Nothing is saved without your confirmation of the preview.
  • A backup before every save; restore any of them.
  • The server refuses to write a module whose logic it does not fully understand, or one changed since the preview.
  • Each account sees only its own installations; administrators approve accounts and extra installations.