MCP or the Model Context Protocol is a method for services to integrate into Large Language Models such as Copilot, Gemini, Claude and ChatGPT.
It provides a range of tools that the AI can call to get information about assets and data.
With these tools, you can directly ask LLMs in plain language to produce reports, analytics and detailed descriptions of the various parts of your system.
We have a public MCP server for our Paint Line demo.
You can try out MCP using Claude.
Firstly, you'll need to install and enable the MCP Addon. This gives your ARDI server dedicated functions to give data to AI models.
You can fill in some details about your system and data sources in the settings, available from Administration | MCP Settings once the addon is enabled.
If you are planning on using a web-based MCP server, you'll need to install the MCP service. This is available as a Python module. See the service install instructions.
Most ARDI servers run on local networks and aren't exposed to the Internet. Because of this, you won't be able to use web-based chat to talk to your ARDI system, as there's no way for the provider (OpenAI, Microsoft, Anthropic etc.) to reach your server to ask for information.
In most cases, you'll use a desktop application such as Claude Desktop. This allows your computer to work as a 'bridge' between the two networks - able to reach out to the Internet to talk to the LLM, but also able to reach the ARDI server to ask for data.
The method to connect varies between applications.
Microsoft Copilot (Coming Soon)
We will include additional instructions as products become available. If you'd like instructions for your preferred LLM tool - or would like to share instructions you've created - please contact us.
Since most ARDI services are on local OT networks, we don't suggest giving Internet-facing AI models direct access to your ARDI server to send MCP requests.
However, we have developed a solution that uses an internet-hosted MQTT broker as a 'go-between' that allows you to make an ARDI server that is behind a firewall (but has access to the Internet) accessible to web-based tools.
This lets you add a Copilot Studio agent to Teams and ask questions about assets while minimising the potential risk of exposing information to the wider Internet.
The MCP addon is always being updated to stay in-sync with changes to common models such as Claude and Chat-GPT.
If you encounter issues with tools not responding as expected, please send us a copy of the tool call parameters sent by your agent.
When tool calls fail, the majority of issues are caused by changes in the formatting of the parameters passed to ARDI. For instance, older versions of Claude would pass a relative range as '1 day' or '3 hours', while newer versions will use '-3d/now' and Copilot/Chat-GPT will often send them as '-3d,now'.
If the format chosen by your agent is reasonable, we'll extend our support to include it.