Show IH: I built an open-source MCP gateway to keep AI agents from breaking production systems (21 tools + Free tier)

Hey Indie Hackers 👋

Like many builders experimenting with agentic workflows (LangChain, AutoGen, CrewAI, MCP), I noticed a glaring problem: probabilistic models shouldn't have unrestricted write/execute access to real-world APIs or physical hardware.

When an agent hallucinates a payload or gets stuck in a recursive tool-calling loop, it burns budget and risks production stability.
To solve this, I built SEOSiri Industrial AI Gateway (IAIG).

💡 What is it?
It’s a deterministic control plane that acts as a proxy between your AI agents and your hardware/APIs using the Model Context Protocol (MCP) standard.

🛠️ What’s included:
Tool Inspection & Firewall: iaig_firewall_inspect validates payloads before execution.
Rate & Loop Limiting: iaig_budget_gate & iaig_throttle_tool_context prevent runaway costs.
Human-in-the-Loop: iaig_queue_proposal_hmi for high-risk manual confirmations.
Auditability: iaig_audit_ledger for immutable operational records.

📦 Details:
License: MIT (Open Source)
Package: Published on npm
Pricing: Free tier live (no credit card required)

🔗 Link to Docs & API Access:
https://www.seosiri.com/2026/08/industrial-ai-api.html

I'd love to hear your feedback on the architecture, tool coverage, and what other safety gates you look for when deploying agents!