One protocol.
Every device.

NRP standardizes how AI systems observe, control, and enforce safety constraints on any device — from servers to robots to industrial PLCs.

View Spec →pip install nrprotocol

$ pip install nrprotocol Installed nrprotocol-0.1.0 $ python3 >>> from nrp import NRPId, NRPDriver, NRPManifest >>> NRPId.parse("nrp://factory/robot/arm-7") NRPId(scope='factory', kind='robot', name='arm-7')

Interface

Three operations. Four-method contract.

Every device implements observe, act, shield. The control plane reads the manifest and generates tools.

01 OBSERVE

Read state

Typed channels with units. Sensors, metrics, feeds, API state.

02 ACT

Execute commands

Shell, movement, writes, API calls. Audited and constrained.

03 SHIELD

Enforce constraints

Device-declared safety limits. Speed, workspace, transfer caps.

04 MANIFEST

Self-description

Nodes declare capabilities at registration. Tools auto-generated.

05 EVENTS

Async push

Severity-routed. Emergency bypass — under 100ms dispatch.

Drivers

12 protocols covered.

SSH

Servers

HTTP Auto

OpenAPI/GraphQL

MQTT

IoT pub/sub

WebSocket

Realtime

Modbus

RTU/TCP PLCs

OPC-UA

SCADA

ROS 2

Robots

Unitree

Humanoids

Docker

Containers

Chrome CDP

Browser

DDS

Telemetry

Raw Socket

TCP/UDP

Code

Four methods. Any device.

from nrp import NRPDriver, NRPManifest, ChannelSpec, ActionSpec class Sensor(NRPDriver): def manifest(self): return NRPManifest(nrp_id=self._nrp_id, observe=[ChannelSpec("temperature", "float", unit="°C")], act=[ActionSpec("set_threshold", {"value": "float"})]) async def observe(self, ch=None): return {"temperature": self.read()} async def act(self, cmd, args): self.threshold = args["value"] def shield_rules(self): return [ShieldRule("max", LIMIT, 85.0)]

Comparison

NRP extends MCP to hardware.

	MCP	NRP
Software integration	✓	✓
Hardware control	—	✓
Safety constraints	—	✓
Self-describing devices	—	✓
Push events	—	✓
Device identity	—	✓
Auth levels	—	✓
Tamper-evident audit	—	✓

One protocol.Every device.