MCP Tools server endpoint for compatible AI agents

Overview

SorbaAI (SORBA.ai) is an industrial Artificial Intelligence platform focused on AutoML (automated Machine Learning), designed to simplify the creation and operation of predictive and prescriptive models using process and asset data. In practical terms, it helps companies turn data into insights and forecasts to improve equipment reliability, optimize operations, and support real-time decision-making.

This connector is a TagProvider Extension Module. This means it can also be developed by anyone.

In this page you will see information related to how to configure, what it creates, and how to implement and utilize this connector.

TagProvider Services Settings

PrimaryStation

In Unified Namespace / TagProvider Services, in the PrimaryStation column, you will find this settings:

IP: IP address of the SorbaAI server.

Algorithm: Machine learning algorithm used to train and run the model. Supported algorithms:

• Regression
• DigitalTwin
• Optimization
• Clustering
• Forecasting
• Classification

DatabasePath: Path to the dataset used for initial training. Currently, only .csv files are supported. In future releases, it will be possible to train using tables from a database connection.

Asset Tree

After creating the Unified Namespace / TagProvider Services connection, link it to a folder in the Unified Namespace / Asset Tree. If you don’t already have a folder created, create a new one.

Once the connection is linked to an Asset Tree folder, open the folder and the available options will depend on the selected algorithm. In general, you will see the Asset Tree structure below:

Folder
+-- ConnectionAlias
    +-- CONFIG
    |   +-- RUNTIME
    |   |   +-- autoadjust_mode
    |   |   +-- autoadjust_start
    |   |   +-- autoadjust_time_window(min)
    |   +-- TRAINING
    |       +-- finish_training_date
    |       +-- model
    |       +-- start_training_date
    |       +-- training_begin
    |       +-- training_log_message
    +-- INPUTS
    +-- OUTPUTS
    +-- TARGET
        +-- response_variable

Properties from the Asset tree explained in the table below:

Workflow

1) Configuration

The customer must have the SorbaAI service running (locally, remotely, or on a virtual machine). In Unified Namespace / TagProvider Services / PrimaryStation, the customer configures the connection by:

• Entering the SorbaAI service IP address

• Selecting the model to be used

• Defining the training dataset file

2) Model Training

After the configuration is complete, the customer links the SorbaAI connection to a folder in the Unified Namespace / Asset Tree. Once linked:

• The required subfolders are created automatically (as described in the Asset Tree section).

• When the training_begin property is triggered, the system automatically creates the properties under:

  • INPUTS

  • OUTPUTS

  • TARGET

These properties are also created as tags in the solution, enabling:

• Real-time input of values (INPUTS)

• Real-time access to model results (OUTPUTS and TARGET)

3) Model Usage

Once the tags under INPUTS, OUTPUTS, and TARGET are available:

• The application writes or maps real-time field values to INPUTS

• The model produces real-time results through OUTPUTS and TARGET tags

• These results can be used throughout the solution, for example:

  • Displaying values on screens/dashboards

  • Creating alarms and notifications

  • Driving logic or automation based on model outputs

In this section...