. N3uron Academy | Introducing MCP Server
Contact Us
Download N3uron
Back to videos

MCP Server / Introducing MCP Server

Introducing MCP Server

Description

In this video of our N3uron Academy, we’ll explore how N3uron can expose industrial context to AI clients through MCP, or Model Context Protocol. Let’s get started!

  • [09:29] Introducing MCP Server

  • [11:09] Configuring MCP Server

Transcription

[00:00] Welcome to this new N3uron Academy series focused on the N3uron MCP Server. In this series, we are going to explore how N3uron can expose industrial context to AI clients through MCP, or Model Context Protocol. Before we start configuring anything, it is important to understand the core idea behind MCP and why it matters in this type of workflow. At a high level, MCP is a standard that allows AI applications to discover and use external context, tools, prompts, and resources through a consistent interface. It has become the de facto standard for connecting LLM-based clients to external systems and APIs. It is best understood as a layer that complements traditional APIs, rather than replacing them. In a traditional API, a human programmer reads documentation, understands the available functions, and then writes the code needed to interact with them.

[01:03] MCP brings a similar idea into the world of LLMs by making capabilities easier to expose in a way that AI clients can understand and use more dynamically. In other words, what APIs have traditionally been for human programmers, MCP is for LLM-based clients. It gives models a more consistent way to reach external systems without requiring a different custom integration every time. It also means that N3uron MCP Server can be used alongside other MCP servers, all through the same underlying protocol. So instead of treating every external system as a separate integration problem, an AI client can interact with multiple MCP-enabled systems in a much more unified way. And that matters because a language model on its own does not directly interact with plant data, alarms, historians, or operational structures. It can generate responses, but it still needs a structured way to access real systems and work with live operational context.

[02:00] That is exactly where MCP becomes useful. This also introduces the idea of an agent. In this series, when we talk about an agent, we mean the runtime that orchestrates the full decision loop around the model. First, it builds the context and sends it to the LLM. Then the LLM decides the next action. Next, the agent executes the corresponding tools through MCP. The results are then used to update the context, and the loop repeats until the goal is achieved. In a practical sense, you can think of the agent as the layer that sets the game around the model. It defines the rules, the limits, the available tools, the permissions, the guardrails, and the way context is managed during the interaction. The LLM then plays the next move inside that environment. So even though N3uron MCP Server already exposes governed capabilities from the N3uron side, the agent still adds its own layer of control through context management, tool management, skills, and execution logic.

[03:00] This is why agents matter so much in industrial AI systems. An LLM by itself does not have real-time plant data, tool execution, or durable operational context. The agent runtime is what connects the model to those external capabilities and turns reasoning into action. Depending on the use case, that agent could be Claude Desktop, Codex, LibreChat, ChatGPT, Antigravity, n8n used as an agent, or other similar clients. In our case, we will use Claude Desktop as the main agent and MCP client, mainly because it is simple to use and gives us a very practical way to start working with N3uron MCP Server. Now that we have seen what MCP is and what the agent does, the next question is where N3uron MCP Server fits in that architecture. N3uron acts as the industrial context layer between the AI client and the plant, exposing the three main MCP primitives out of the box: tools, prompts, and resources. On the tools side, it provides more than forty built-in tools across namespaces such as Tags, Alarms, Modules, Logs, System, Backup, and more, and those capabilities can be extended further with custom tools implemented through the Scripting module.

[04:11] Prompts can also be exposed as reusable, parameterized instructions for industrial workflows, and resources can publish operational content such as tags, alarms, logs, and embedded files. Some MCP clients are still mainly tool-centric, but N3uron MCP Server supports the protocol natively and is ready to expose all three primitives through the same governed interface. Access is controlled by bearer tokens, scoped permissions, and tag-model filters, so each client only discovers and uses what it is allowed to access. For the hands-on part of this series, we are not going to start from an empty project, and we are not going to require a direct connection to a real plant. Instead, we will use a downloadable backup of the PV Demo project so viewers can follow along with the same environment used in these videos. This makes the whole series much more accessible because you do not need PLCs, field devices, or a production system to begin learning and experimenting with MCP Server.

[05:05] You can simply restore the project and start exploring. And of course, if you already have your own project or your own architecture, you are completely free to adapt everything we explain here to your own use case. As you can see on screen, this backup project is a complete, end-to-end simulation of a photovoltaic plant. It includes a full WebVision HMI designed for plant operators, with everything you’d expect in a real deployment: overview screens, alarms, real-time data values, heatmaps, weather-station information and forecasts, and historical trends for deeper analysis, etc. We’ll also use the WebUI interface to analyze tag model, logs, real-time values, and historical trends alongside the HMI, so we can validate the AI agent responses. This gives us a practical way to test requests, confirm results, and quickly verify that the agents behave as expected. Before moving on, I also recommend requesting a 30-day trial license.

[06:02] That will give you more time to work comfortably during a proof of concept. In demo mode, N3uron allows each module to run for two hours at a time. That period can be restarted as many times as needed, but a trial license gives you a much smoother experience when you want to test, modify the project, and spend more time validating ideas without interruptions. Now, let’s take a brief look at the structure of the project we will be using throughout the series. This is a comprehensive and self-contained simulation of a photovoltaic plant. If we open the project and browse the tag model, we can already see that it contains a meaningful industrial hierarchy, with plant areas, equipment, KPIs, and related structures that represent a realistic PV environment. This is important because we are not working with a few disconnected demo variables. We are working with a project that already has enough structure and context to make MCP interactions useful from the very beginning. The project initially is configured with the local embedded MongoDB database.

[07:00] That will give you more time to work comfortably during a proof of concept. In demo mode, N3uron allows each module to run for two hours at a time. That period can be restarted as many times as needed, but a trial license gives you a much smoother experience when you want to test, modify the project, and spend more time validating ideas without interruptions. Now, let’s take a brief look at the structure of the project we will be using throughout the series. This is a comprehensive and self-contained simulation of a photovoltaic plant. If we open the project and browse the tag model, we can already see that it contains a meaningful industrial hierarchy, with plant areas, equipment, KPIs, and related structures that represent a realistic PV environment. This is important because we are not working with a few disconnected demo variables. We are working with a project that already has enough structure and context to make MCP interactions useful from the very beginning. The project initially is configured with the local embedded MongoDB database.

[08:04] It allows the same software base to be used across very different environments, from small embedded devices all the way to larger server infrastructures. And because configuration and visualization are already integrated into the platform, we do not need external tools just to get started. Of course, you do not need to use the same hardware I am using. You are free to deploy N3uron and this project on any architecture and operating system supported by N3uron. That includes Windows and Linux installations, ARM-based devices such as Raspberry Pi, containerized deployments with Docker or Kubernetes, and cloud environments such as AWS. This flexibility is one of the strengths of the platform, because the same approach can be used from lightweight edge devices all the way to larger server or cloud deployments. And if you run into any issues while installing the software or trying to follow the tutorial, do not hesitate to contact N3uron support. So in this first video, the goal is simply to establish the foundation for everything that comes next.

[09:02] We have introduced what MCP is, we have clarified what we mean by an agent, we have explained why we will use Claude Desktop as our main MCP client, and we have presented the PV Demo project that will serve as the base for the rest of the series. In the next video, we will move into the configuration side and start looking in more detail at the connection process and the first interactions with our N3uron node.

Contact Us
Download N3uron