Skip to main content
Governed AI access for industrial operations

Proxus MCP Server

Give approved AI agents the operational context to answer with evidence: current device health, configuration, dependency paths, telemetry history, and permitted business actions. Keep identity, scope, and control boundaries inside Proxus.

20
governed tools
Role scoped
existing user permissions
Bounded
graph and telemetry work
No direct OT
control session exposure
Industrial operations report produced from Proxus MCP evidence
REAL MCP OUTPUT

Operational evidence an agent can actually produce

This report was produced from governed Proxus context. The visual remains unmasked so teams can inspect the findings, evidence boundaries, and operational summary rather than seeing an abstract AI illustration.

Open full-size report

Compatible with leading AI models

OpenAI
OpenAI
Anthropic
Anthropic
Google
Google Gemini
Meta
Meta Llama
Microsoft
Microsoft Copilot
Hugging Face
Hugging Face
OpenAI
OpenAI
Anthropic
Anthropic
Google
Google Gemini
Meta
Meta Llama
Microsoft
Microsoft Copilot
Hugging Face
Hugging Face
OpenAI
OpenAI
Anthropic
Anthropic
Google
Google Gemini
Meta
Meta Llama
Microsoft
Microsoft Copilot
Hugging Face
Hugging Face
CONTEXT BEFORE ACTION

An agent arrives knowing how to learn Proxus

The server publishes its authentication workflow, tool contracts, entity schema, evidence semantics, limits, and recovery actions. Users do not have to write a perfect master prompt for every assistant.

Discover

Live tool schemas and platform overview route the agent to current device status, graph, telemetry, logs, or governed entities.

Authorize

The connected agent operates as the selected Proxus user. Existing role, object, field, gateway, and UNS scope remain authoritative.

Investigate

Bounded dependency and telemetry workflows keep evidence, freshness, redactions, omitted scope, and resource limits visible.

Recover

Structured errors tell the agent which input, permission, catalog refresh, or narrower query is required instead of ending the workflow with a generic failure.

OPERATIONAL INTELLIGENCE GRAPH

Trace what feeds what, and why

Proxus connects devices, profiles, tags, rules, functions, dashboards, alerts, gateways, and telemetry evidence into a bounded operational graph. Relationships keep their source, confidence, and redaction context.

Follow the graph workflow
Impact

Before a profile, tag, rule, or function changes, find the downstream consumers that may be affected.

Lineage

Follow an alert, dashboard, or function upstream to the configuration and telemetry semantics that feed it.

Causal paths

Rank cycle-safe paths for a specific maintenance, runtime, or evidence question without treating bounded scope as proof of absence.

Edge evidence

See whether a relationship comes from typed configuration, desired runtime state, parsed rule, dashboard, or function semantics.

ANSWERS WITH COVERAGE

From operational question to verifiable evidence

What is online now?

Read runtime freshness and gateway evidence for the devices the current user can see. Keep configuration enabled state separate from live health.

What will this change affect?

Walk downstream dependencies, preserve confidence and source, and expose when the graph is bounded, redacted, or unable to prove a dynamic relationship.

What happened at the alarm?

Trace the alert to its rule, device, and telemetry key, then inspect a narrow time window to compare the measured value with the recorded event.

GOVERNANCE IN THE REQUEST PATH

AI access does not bypass Proxus security

A protected call is resolved to one Proxus identity, checked against existing permissions, bounded by the tool contract, and audited. Database credentials and control-system sessions are not handed to the agent.

One identity boundary. The protocol-bound user remains the downstream identity; ambient headers cannot replace it.
Permission-scoped results. Objects, fields, device status, graph snapshots, logs, and operations respect the current role.
Guarded telemetry. Administrator-only read queries stay inside the approved telemetry surface and finite resource limits.
Explicit writes. A vague instruction cannot become a mutation; the exact human request, target, method, minimal body, and permissions must align.

Give agents context without giving up control

Start with a governed read-only deployment, validate permission boundaries with real restricted users, then enable only the business operations your teams approve.