The SEOSiri Industrial AI Gateway (IAIG) is an open-source, MIT-licensed zero-trust MCP server that lets AI agents operate SCADA/MES systems, ISA-95 Unified Namespaces, digital twins, and ROS 2 robot fleets — without giving those agents raw, unaudited access to physical hardware. Every proposed action moves through 6 architectural layers and 18 discrete MCP tools before it ever reaches a PLC or robot controller. Install with npm install @seosiri/industrial-ai-gateway; production and enterprise access run through the SEOSiri Developer Portal.
Author: ✍️ Momenul Ahmad — Founder & Lead Architect, SEOSiri
What Is the SEOSiri Industrial AI Gateway?
The SEOSiri Industrial AI Gateway (IAIG) is an open-source Model Context Protocol (MCP) server — a standardized interface that lets large language model agents call external tools — purpose-built to bridge AI agents and industrial cyber-physical systems (equipment where a software decision directly triggers a physical action). It translates ISA-95 Unified Namespace data, bridges SCADA/MES protocols (Modbus TCP, OPC UA, Sparkplug B), runs every proposed action through a physics-aware digital twin — a live simulated replica of the machine — before it reaches real hardware, and authorizes ROS 2 robot fleet commands, signing each validated instruction with a hardware-bound TPM 2.0 signature.
📌 Key Takeaways
- Six decoupled layers separate AI reasoning from physical actuation, so a hallucinated instruction can never directly move a robot arm or trip a PLC.
- 18 MCP tools span connectivity, telemetry compression, safety validation, hybrid reasoning, access control, and fleet robotics.
- Deterministic safety, not probabilistic trust: every action is simulated against a digital twin and checked against hard interlocks before it executes.
- Built for OT/IT convergence: operational technology (SCADA, PLCs, robot controllers) and IT-side AI agents talk through one governed layer instead of ad-hoc integrations.
- Open-source core, commercial edge access: MIT-licensed locally; Pro and Enterprise API tiers unlock production throughput and dedicated infrastructure.
Why Industrial AI Needs a Zero-Trust Gateway
Connecting an autonomous agent directly to a PLC, robot controller, or SCADA historian changes the failure mode entirely. A bad API call to a CRM produces a support ticket; a bad G-code instruction to an uncalibrated industrial arm can produce physical damage or a safety incident. This is the core tension of Industry 4.0 and IIoT (Industrial Internet of Things) deployments: OT networks are built around determinism and physical safety margins, while LLM agents are probabilistic by design.
IAIG resolves that mismatch by applying zero-trust principles — verify every action, trust nothing by default, similar in spirit to the architecture described in NIST SP 800-207 — to the specific problem of agent-to-machine command execution, aligning with the segmentation and least-privilege practices industrial security standards like IEC 62443 call for on OT networks.
The 6 Core Architectural Layers
| Layer | Function |
|---|---|
| 1. Core Integration & Connectivity | ISA-95 Unified Namespace translation; Modbus TCP, OPC UA, and Sparkplug B protocol bridges. |
| 2. Telemetry Optimization & Token Management | Local ML-driven vectorized summarization and time-series compression, keeping high-frequency sensor streams inside LLM context limits. |
| 3. Deterministic Guardrail Validation Core | Physics-aware digital twin simulation, mechanical hard interlocks, exception feedback loops. |
| 4. Hierarchical Reasoning & Hybrid Routing | Air-gapped Small Language Model (SLM) diagnostics with multi-tier cloud escalation. |
| 5. Security, Access Control & AI Firewalls | Protocol-level prompt inspection and hardware-bound TPM 2.0 command signatures. |
| 6. Advanced Fleet Robotics & IoT Control | ROS 2 Action Goal synthesis (nav2_msgs), E-Nose olfactory arrays, biometric IoT authentication. |
How It Works: The Request Lifecycle
The six layers aren't parallel modules — they're a sequential pipeline every action passes through, in this order:
- Ingest: the agent's proposed action arrives, and Layer 1 translates it against the plant's ISA-95 Unified Namespace and the relevant protocol bridge (Modbus TCP, OPC UA, or Sparkplug B).
- Compress: Layer 2 summarizes and compresses any telemetry context the agent needs to reason over, so high-frequency sensor data doesn't blow the model's context window.
- Validate: Layer 3 runs the proposed action through the digital twin simulation and mechanical hard interlocks. Anything that fails is rejected here and routed back to the agent as structured feedback — nothing physical happens yet.
- Reason: Layer 4 decides whether the decision can be resolved locally by an air-gapped SLM or needs cloud-tier escalation for more complex reasoning.
- Sign: Layer 5's AI firewall inspects the final instruction for prompt-injection or policy violations, then signs it with a hardware-bound TPM 2.0 key.
- Execute: only a signed, validated instruction reaches Layer 6, which dispatches it to the ROS 2 fleet, PLC, or IoT actuator.
An unsafe instruction can be stopped at any of steps 3, 4, or 5 — it never has to reach real hardware to fail safely.
The 18 MCP Tools, at a Glance
Each architectural layer exposes three callable MCP tools to the connected agent:
| # | Tool | Layer |
|---|---|---|
| 1 | ISA-95 UNS Translator | Connectivity |
| 2 | Modbus TCP / OPC UA Bridge | Connectivity |
| 3 | Sparkplug B Bridge | Connectivity |
| 4 | Telemetry Vectorized Summarizer | Telemetry |
| 5 | Time-Series Compression Buffer | Telemetry |
| 6 | Token Budget Manager | Telemetry |
| 7 | Digital Twin Simulator | Guardrail Core |
| 8 | Mechanical Hard Interlock Checker | Guardrail Core |
| 9 | Exception Feedback Router | Guardrail Core |
| 10 | Air-Gapped SLM Diagnostic Executor | Reasoning |
| 11 | Multi-Tier Cloud Escalation Router | Reasoning |
| 12 | Reasoning Tier Selector | Reasoning |
| 13 | AI Firewall / Prompt Inspector | Security |
| 14 | TPM 2.0 Command Signer | Security |
| 15 | Access Control Policy Engine | Security |
| 16 | ROS 2 Action Goal Synthesizer | Fleet Robotics |
| 17 | E-Nose Olfactory Array Reader | Fleet Robotics |
| 18 | Biometric IoT Auth Bridge | Fleet Robotics |
Tool names above are mapped from IAIG's six published architectural layers (3 tools/layer). Confirm exact tool identifiers against src/tools/ in the GitHub repository before quoting them in integration docs.
Who Should Use This
IAIG targets teams connecting AI agents to physical operations rather than pure software workflows:
- Manufacturers and system integrators running SCADA/MES lines who want agent-assisted diagnostics without exposing PLCs directly to an LLM.
- Warehouse and logistics robotics teams orchestrating ROS 2 fleets where a bad navigation goal has physical consequences.
- Energy and utilities operators bridging OPC UA/Sparkplug B telemetry into agentic monitoring and predictive-maintenance workflows.
- Industrial software vendors embedding agentic features into existing OT tooling without building a safety-validation layer from scratch.
IAIG vs. General-Purpose AI Gateways
"AI gateway" usually refers to LLM traffic routers — tools that load-balance API calls across OpenAI, Anthropic, or Gemini and add caching, cost tracking, and observability. IAIG solves a different problem: it doesn't route between language models, it routes a model's decisions safely into physical machinery. A generic AI gateway assumes the worst outcome of a bad response is a bad chat reply; IAIG assumes the worst outcome is a robot arm moving somewhere it shouldn't, and is architected around that assumption from the guardrail layer up.
Requirements & Quickstart
IAIG installs like any standard npm package and drops into Claude Desktop or any MCP-compatible client. A Node.js runtime and network access to your ISA-95/SCADA endpoints (Modbus TCP, OPC UA, or Sparkplug B) are the only prerequisites.
npm install @seosiri/industrial-ai-gateway npm run build npm test
Claude Desktop configuration (claude_desktop_config.json):
{
"mcpServers": {
"seosiri-iaig": {
"command": "npx",
"args": ["-y", "@seosiri/industrial-ai-gateway"]
}
}
}
Full source and the MIT license are on GitHub at SEOSiri-Official/industrial-ai-gateway; the package is published to npm as @seosiri/industrial-ai-gateway.
Commercial Licensing & High-Throughput API Access
The core gateway is free and open-source. Production edge-gateway access, dedicated infrastructure, and bespoke plant integration run through three tiers on the Developer Portal:
| Tier | Access | Best For |
|---|---|---|
| Free | 30 requests/minute | Prototyping, evaluation, single-agent testing. |
| Pro — $299/mo | 1,000 req/min across all industrial edge gateways | Production deployments running continuous agent-to-plant workloads. |
| Enterprise — from $2,500 | Dedicated Zero Trust VPC, custom PLC drivers, on-site TPM calibration | Manufacturers and integrators running regulated, multi-site operations. |
Ready to scale beyond the free tier? Request Pro or Enterprise API keys, custom PLC driver support, or a Zero Trust VPC deployment through the SEOSiri Developer Portal, or reach the systems architecture team at [email protected].
Part of the SEOSiri Sovereign MCP Ecosystem
IAIG interoperates with SEOSiri's broader open-source MCP stack: the Biorobotics Actuation Core shares its physical-safety philosophy at lab scale; the API Guard MCP Server uses the same signed policy-plane pattern as IAIG's guardrail layer; and the Biometric IoT Bridge supplies the mobile authentication IAIG's fleet-robotics module authorizes against. The full catalog of 15+ sovereign MCP servers is indexed at the SEOSiri MCP Ecosystem hub.
Frequently Asked Questions
What is the SEOSiri Industrial AI Gateway used for?
It lets AI agents safely interact with SCADA/MES systems, ISA-95 Unified Namespace data, digital twins, and ROS 2 robot fleets, validating every proposed action against physics-aware safety guardrails before it reaches real hardware.
How does IAIG prevent unsafe robot or PLC commands?
Its Deterministic Guardrail Validation Core runs a physics-aware digital twin simulation and mechanical hard interlocks against every proposed action, and only signs trajectories that pass validation with a hardware-bound TPM 2.0 signature.
What industrial protocols does the gateway support?
Modbus TCP, OPC UA, and Sparkplug B, bridged through an ISA-95 Unified Namespace translation layer.
How is IAIG different from a general AI gateway?
General AI gateways route and load-balance LLM API traffic. IAIG routes validated agent decisions into physical machinery, adding a digital-twin safety simulation and hardware-signed command execution that generic gateways don't provide.
How do I get production API access to the Industrial AI Gateway?
Free tier access is available at 30 requests per minute. Production Pro and Enterprise tiers, including dedicated Zero Trust VPC deployments, are available through the SEOSiri Developer Portal at developers.seosiri.com.
⚠️ Safety & Regulatory Note
IAIG and its guardrail simulations are reference implementations suitable for prototyping, integration testing, and controlled industrial pilots. Deployments governing safety-critical machinery, regulated manufacturing lines, or human-proximate robotics require independent engineering and regulatory validation beyond the scope of this open-source repository.
