Hi all,
I'd like to ask COPA-DATA, and anyone in the community who has tried this, about plans for first-class LLM agent integration with zenon.
AI agents (Claude, ChatGPT, Copilot and others) are quickly becoming part of everyday engineering and operations work. The emerging standard for connecting them to external systems is the Model Context Protocol (MCP), alongside vendor-specific options such as Claude Skills and plain REST APIs with tool calling. Several other industrial software vendors have already started publishing MCP servers for their platforms.
Today, the options I'm aware of for connecting an agent to zenon are all indirect:
- The IIoT Services REST API, which requires the IIoT Services stack and is mainly oriented around variable and archive data.
- Process Gateway (OPC UA, MQTT etc.), which needs a custom bridge between the agent and the protocol.
- The Add-In framework or COM API, which means writing and maintaining your own service inside the Service Engine.
These all work, but each one means building and securing a custom integration layer yourself.
My questions:
1. Does COPA-DATA have plans for an official MCP server, or similar agent tooling, for the Service Engine or IIoT Services? If so, is there a rough timeline or target version?
2. Is a native REST (or similar) endpoint directly on the Service Engine being considered? I mean something lightweight that lets an agent read variables, alarms (AML), events (CEL) and archive data without the full IIoT Services deployment.
3. How would authorisation be handled? I'd want read-only by default, with any write or command access tied into zenon user administration, authorisation levels and the audit trail, so agent actions are traceable in the CEL.
4. Is anything planned on the Engineering Studio side, such as agent-assisted project configuration, bulk variable engineering, or project review through the Engineering Studio API?
5. In the meantime, is there a recommended or supported pattern for doing this with the current product?
Example use cases:
- Natural-language queries over live and historical data ("what tripped feeder 3 last night, and what alarms preceded it?").
- Alarm and event summarisation for shift handovers and incident reports.
- Assisted diagnostics that correlate CEL/AML entries with archive trends.
- Engineering QA, such as checking variable naming, limits and driver configuration against a project standard.
I appreciate that anything touching control actions in an OT environment needs to be approached carefully. A read-only, well-secured interface would already cover most of these use cases.
Has anyone in the community already built something along these lines? I'd be interested to hear what approach you took and what the pitfalls were.
Thanks in advance!