SINT Protocol integration: MCP-layer authorization for Partenit physical AI agents

[pshkv]

Hi Partenit team 👋

Found your project while mapping the physical AI safety ecosystem. The Isaac Sim + ROS2 stack you've built is exactly the deployment environment SINT Protocol was designed to secure at the MCP layer.

The gap we both address:
Physical AI agents (warehouse robots, autonomous vehicles, surgical systems) need:

1. Cognitive safety — your domain (cognitive control infrastructure)
2. Authorization safety — our domain (which tool calls are permitted, at what tier, by whom)

What SINT adds to a Partenit deployment:

When your Partenit agents call MCP tools, SINT's PolicyGateway enforces:

• T0: read-only sensor queries → auto-allow
• T1: path planning queries → allow + audit
• T2: movement commands → async human approval (10s timeout, then deny-safe)
• T3: hardware state changes (e-stop, actuator override) → synchronous human sign-off required

The T2/T3 tiers map directly to your "human-in-the-loop for physical world actions" requirement.

We have a ROS2 bridge (packages/bridge-ros2/) that wraps ROS2 service calls with SINT capability token enforcement. Could be a natural complement to Partenit's safety layer.

Question: Are you using MCP as a tool-call interface for your agents, or a different protocol? Happy to discuss integration architecture.

Repo: https://github.com/sint-ai/sint-protocol (MIT, TypeScript + Python SDK, 1,105 tests)