Product · Agents and MCP
An MCP server for industrial data.
An MCP server lets an AI assistant use another system's tools and data. MaestroHub's MCP server gives assistants like Claude governed access to your factory: named values with their meaning, inside a person's permissions, with every call audited.
In one picture
The assistant never talks to the plant directly.
Coming soonMaestro agent
What an agent can do
With the tools in release 3.0.
Find and read the plant
Search the namespace in plain words, describe a topic, and read live and recent values with their unit, quality and where they came from.
uns_semantic_find_topics, uns_describe_topic, query_topic_data
Watch what is happening
Check the health of connections, the system and alarms, so an agent can say what is broken before it suggests a fix.
monitor_connections, get_system_status, manage_alerts
Build, not only read
Create and change pipelines, dashboards and connections, inside the same permissions a person would have.
manage_pipelines, manage_dashboards, manage_connections
Why it is safe
Governed by the same rules as a person.
It acts as a named person
An agent connects with a credential that belongs to a person. It can see and do only what that credential allows, and every call is recorded under it.
Each agent has its own permissions
Permissions sit on the credential, not only on the user, so two agents of the same person can hold different rights. Some rights, reading the audit log among them, can never be given to an agent.
Every call is in the audit trail
Reads, changes and refusals land in the hash-chained, tamper-evident audit trail, with the person the agent acted for.
Writes to machines stay governed
A write back to a machine or a business system goes through the same permissions and audit as a person's. An approval step for agent actions holds a write until a person approves it.
Questions
What is an MCP server for industrial data?
An MCP server is a standard way for an AI assistant, such as Claude, to use tools and data from another system. An MCP server for industrial data gives the assistant governed access to factory data: it can find machines and signals, read live and recent values with their meaning, and, if allowed, build or change things, instead of guessing from raw tags.
What is the safest way to connect Claude or ChatGPT to industrial systems?
Do not connect the assistant to PLCs or business systems directly. Put a layer in between that holds the connections, gives every value its meaning, and checks every request against a named person's permissions, with an audit trail. That is what MaestroHub's MCP server does: the assistant talks to MaestroHub, and MaestroHub talks to the plant.
How can an LLM read live PLC data?
Connect the PLC to a platform that speaks its protocol, such as OPC UA, Siemens S7 or Modbus, and expose the data to the model through an MCP server. With MaestroHub the model reads named values like "Pump 1, vibration, 3.4 mm/s, good" rather than a raw register, and only the values its person may see.
How do I connect Ignition to an AI agent?
MaestroHub reads from Ignition through its connector and exposes that data, together with your other machines and systems, through its MCP server. Ignition keeps running the plant floor; the agent gets context from all sources in one place, under one permission model.
Which AI clients work with it?
Any client that supports the Model Context Protocol, such as Claude Desktop, Claude Code or Cursor. The MCP server is live in release 3.0.
Does the agent need internet access to my plant?
No. MaestroHub runs on your own infrastructure. Where the AI client runs, and which model it uses, is your choice.
Let your AI client try it.
Download the free trial, point Claude or any MCP client at it, and connect the Digital Factory Simulator if you have no plant to hand.