Making your Data Available to AI Agents

In this section, we will describe:

  • how ScaleOut exposes data to AI agents such as Claude Desktop,

  • how to use metadata in your digital twin models to help AI understand your data

  • how to use AI agents such as Claude Desktop to make sense of live data

Exposing Data: ScaleOut MCP Server

ScaleOut provides an MCP (Model Context Protocol) server that exposes the live data held in your digital twins to AI agents, such as Claude. Rather than requiring an agent to be custom-built or fine-tuned for your data, the MCP server acts as a standard bridge: any MCP-compatible agent can connect to it and immediately gain the ability to explore your digital twin models and reason about their current state.

One benefit of using MCP is that the agent will automatically discover and learn about the tools ScaleOut exposes, as it connects to the MCP server. It doesn’t need to be trained.

The following section focuses on what the server’s tools do and why they’re useful. For instructions on connecting Claude Desktop to the ScaleOut MCP server, see Configuring Claude Desktop.

Why These Tools Matter for an AI Agent

Digital twins represent live, constantly-changing data: sensor readings, equipment status, operational metrics, and so on. An AI agent has no built-in knowledge of your specific models, their structure, or their current values. That context has to be discovered at conversation time.

The MCP server’s tools are designed to let an agent do exactly that, in a natural progression:

  1. Discover what models exist and understand their structure.

  2. Look at real data to ground its understanding in concrete examples.

  3. Retrieve the specific instances or data points to further understand the data or respond to a user’s question.

  4. Summarize and visualize that data at an aggregate level, when a trend or comparison matters more than individual records.

With these tools, an agent can independently look things up, verify its assumptions, and answer questions grounded in the live state of your digital twins.

Categories of Tools

Discovery Tools

Tools: list_model_names, describe_model, get_sample_instances

Before an agent can ask a meaningful question about your data, it needs to know what models are available. The discovery tools list the digital twin models available on the server and inspect each model’s schema, its properties and their data types.

It lets the agent adapt to your environment rather than guessing. An agent working with a fleet-management deployment and one working with a manufacturing deployment will discover entirely different models and properties, and can tailor its questions and queries accordingly, without any of that structure having been hard-coded in advance.

Knowing a model’s schema is not the same as knowing what its data actually looks like. The discovery tools also let the agent pull a small set of real, representative instances of a model.

Retrieval Tools

Tools: get_instance, query_instances

Once an agent understands a model, it needs to get at the actual data. The agent looks at the data to further its understanding of the model. It can determine normal operation range, key properties, find outliers or even identify relationships between different properties of digital twins and their operational state.

It can also answer your targeted questions: whether that’s a single, specific twin like “what’s the current status of pump #42?” or a filtered set of twins matching some condition like “which pumps are currently offline?”.

These tools give the agent direct, targeted access to live instance data, so it can answer questions about current, real-world state rather than relying on stale or general knowledge.

Visualization Tools

Tool: create_chart

Some questions are best answered with a summary or a trend rather than a list of individual records (e.g. “what’s the average flow rate by pump status?”). This tool lets the agent aggregate data across many instances (counting, summing, averaging, etc.) and present it as a chart that can be used to monitor over time.

This is valuable because it lets the agent go beyond reporting raw facts at a given time and instead surface the kind of higher-level insight you would otherwise have to build a dashboard or report to get.

Putting It Together

Taken together, these tools let an AI agent connected to the ScaleOut MCP server treat your digital twins as a live, explorable source of truth: it can find out what data exists, ground itself in real examples, retrieve exactly the records that matter, and summarize data at a glance, all within the flow of a natural conversation. Beyond that, it can derive new ways to look at the data by creating its own queries and understand the data at a scale that a human might not.