Breaking down enterprise data silos

In a typical manufacturing environment, your data landscape is fragmented. Product structures live in PLM (like Siemens Teamcenter). Orders and inventory live in ERP (like SAP). Customer interactions live in CRM (like Salesforce). And then there are the Mendix apps built to fill the gaps and digitize processes on the shop floor.

Each system is excellent at what it does. But when a Plant Manager or Quality Engineer asks a cross-domain question like – “Are quality failures correlated with specific shifts, and which machines are responsible?” – IT teams still scramble to stitch together SQL exports, spreadsheets, and one-off integrations.
This is exactly the problem space where Mendix, Altair Graph Studio, and GenAI agents powered by the Model Context Protocol (MCP) form a new, powerful architecture.

To demonstrate how these three technologies converge to move us from static reporting to Agentic AI, we built a comprehensive manufacturing operations application named Evora.

The new stack: Low code, knowledge graphs, and MCP

Before diving into the use case, let’s define the three core ingredients that make this architecture possible.

1. Altair Graph Studio: The semantic fabric

In traditional manufacturing databases, you think in tables and joins. In Altair Graph Studio, you think in entities and relationships. Instead of row 402 matching row 881, the graph understands the business context:

• Assembly→ is composed of → Components
• Machine→ is operated by →Operator
• Defect→ occurred during →Shift

Graph Studio becomes the map of your enterprise, ingesting data from Teamcenter, SAP, and Mendix, and enriching it with an ontology (formal description of how your business concepts relate to one another).

2. Mendix: The orchestration layer

Mendix remains the “front door” and control plane. In this architecture, Mendix isn’t just a UI; it is the environment where the agent lives alongside the action. It hosts the portals where quality engineers review insights and, crucially, triggers the workflows to rectify issues (e.g., creating a maintenance ticket or flagging a batch).

3. MCP: The standard language for agents

This is the game-changer. The Model Context Protocol (MCP) is an open standard that allows you to expose data and tools to Large Language Models (LLMs) in a consistent way. Instead of hard coding an integration between a chatbot and a database, you build an MCP Server. This server acts as a catalog that tells the AI: “Here are the tools I have (e.g.,execute_sparql_query,get_quality_metrics), and here is how you use them”.

The “Evora” scenario: Root cause analysis on the factory floor

Let’s look at this in practice. In our demo app, Evora, we track manufacturing data across multiple factories, including assembly records, chassis data, and operator logs.

The problem

A Quality Engineer notices a dip in metrics but doesn’t know why. Traditionally, finding the root cause would require querying three different systems or asking a data analyst to slice the data by shift, then by machine, then by operator.

The agentic app solution

Embedded directly inside the Mendix application is an AI agent. Because we have exposed the Altair Knowledge Graph via an MCP Server, the engineer can simply ask:
“Are quality failures correlated with specific shifts? Which shift has the highest defect rate?”

What happens under the hood?

This is where the “agentic” part shines. The LLM (e.g., Claude Sonnet or OpenAI GPT) analyzes the intent. It realizes it doesn’t have this answer memorized, but it sees a tool in its MCP toolkit to query the graph.

1. The reasoning: The agent constructs a query to the knowledge graph to fetch defect counts grouped by shift.
2. The execution: The MCP server executes the query against Altair Graph Studio.
3. The insight: The graph returns the raw data, and the agent synthesizes it into a clear answer.

The engineer can then follow up: “Drill down into the Night Shift. Is it a specific machine?” The agent simply calls the necessary graph tools again to refine the answer.

Write once, run anywhere: The power of MCP

Why is this better than a standard API integration? Portability.
Because the logic is wrapped in an MCP Server, the exact same capability used by the Mendix app can be plugged into other AI clients without rewriting a single line of code:

• In Mendix: The shop floor workers use the agent to troubleshoot issues in real-time.
• In Altair Agent Studio: Data scientists use the same graph connection to prototype new analytical flows.
• In ChatGPT/Claude Desktop: An Enterprise Architect can connect to the “Evora Graph MCP” to ask high-level architectural questions about the data model.

You aren’t building a chatbot; you are building a reusable AI-native integration fabric.

Why this resonates with manufacturing IT

For Siemens customers and industrial enterprises, this architecture solves three critical challenges:

1. Context over chaos: It moves beyond searching documents. By using a Knowledge Graph, the AI understands the structure of your BOMs and processes, reducing hallucinations and increasing accuracy.
2. Governance: Because the AI is accessing data through an MCP server, IT controls exactly what the AI can see and do. You aren’t dumping your database into a public LLM; you are giving an agent specific, governed tools.
3. Agility: Mendix provides the speed to build the user interfaces and workflows that turn these AI insights into real-world actions—like pausing a production line or scheduling operator retraining.

The path forward

The Evora demo proves that the future of industrial software isn’t just about collecting data; it’s about synthesizing it. By combining Mendix for the experience, Altair for the semantic context, and MCP for the connectivity, we are entering the era of agentic applications—apps that don’t just display data, but help you understand and act on it.