ClawHub

NORNR MCP Control

MCP Tools

Put policy before paid actions, require approval for risky autonomous actions, and keep a finance-ready audit trail.

Install

openclaw skills install nornr-governance

NORNR MCP Control for OpenClaw

What this skill is for

Use NORNR as the control layer before an OpenClaw skill or local MCP-connected action triggers spend, vendor action or another consequential downstream step.

The shortest correct reading is:

  • one local tool request becomes one NORNR intent
  • NORNR decides whether it is approved, queued or blocked
  • queued work enters named review with context attached
  • the same action still survives into proof, finance packet and audit export later

When to use it

  • Before a skill triggers a purchase, subscription, or vendor-side paid action
  • When an autonomous flow should pause for operator approval before execution
  • When finance or operations need one defensible decision record after the action completes
  • When suspicious prompt instructions or unusual spend should surface as review posture instead of silent execution

Installation

  • python -m pip install -r requirements.txt

Hello world

  1. Set NORNR_API_KEY.
  2. Run: python nornr_governance.py preflight --action purchase --amount-usd 25 --counterparty openai --purpose "Run the paid research action"
  3. If NORNR returns approved, the autonomous action may proceed.
  4. If NORNR returns approval_required or blocked, stop the autonomous flow and route to review.

Copy-paste MCP / desktop config

Use NORNR as the default control layer before consequential local tool execution clears:

  • Claude Desktop: nornr mcp claude-config --server-name nornr --agent-id openclaw-agent
  • Cursor: nornr mcp cursor-config --server-name nornr --agent-id openclaw-agent

Cursor rule

Add this to .cursorrules if Cursor is allowed to use local tools through NORNR:

Use NORNR as the control layer before consequential tool execution.
Do not proceed with a paid, vendor-side or policy-sensitive action until NORNR returns approved or a named operator explicitly approves the queued intent.
Treat queued, blocked, anomalous or review-required posture as a stop state for autonomous execution.

First run

  1. Set NORNR_API_KEY and, if needed, NORNR_AGENT_ID.
  2. Run: python nornr_governance.py preflight --action purchase --amount-usd 25 --counterparty openai --purpose "Run the paid research action"
  3. If NORNR returns approved, the action may proceed.
  4. If NORNR returns approval_required or blocked, stop the autonomous flow and route to review.

Environment

Only this variable is required:

  • NORNR_API_KEY

These are optional:

  • NORNR_BASE_URL (optional, defaults to https://nornr.com)
  • NORNR_AGENT_ID or a stored NORNR login profile

Default policy pack for OpenClaw

Start from mcp-local-tools-guarded when OpenClaw actions can trigger spend, vendor action or policy-sensitive tool execution. The point is not to invent a new OpenClaw policy language. It is to make NORNR the default control layer before autonomous execution.

Recommended API key scope

Minimum action scope for the full skill surface:

  • payments:write
  • workspace:read
  • approvals:write
  • events:read
  • audit:read

Add these if you want the finance-close paths too:

  • reports:read
  • webhooks:read

Dependency provenance

This skill delegates governance decisions to the official NORNR Python SDK, agentpay.

  • Public MCP package repo: https://github.com/NORNR/nornr-mcp-control
  • Install source: requirements.txt
  • Pinned PyPI package: nornr-agentpay==0.1.0
  • Local bridge: nornr_governance.py

Review the pinned SDK revision before use if your environment requires dependency review.

Commands

  • python nornr_governance.py preflight --action purchase --amount-usd 25 --counterparty openai --purpose "Run the paid research action"
  • python nornr_governance.py approvals
  • python nornr_governance.py approve --payment-intent-id pi_123 --comment "Approved after review"
  • python nornr_governance.py reject --payment-intent-id pi_123 --comment "Rejected pending review"
  • python nornr_governance.py anomalies --counterparty openai
  • python nornr_governance.py timeline
  • python nornr_governance.py finance-packet
  • python nornr_governance.py audit-export
  • python nornr_governance.py weekly-review
  • python nornr_governance.py monthly-close --provider quickbooks
  • python nornr_governance.py review-bundle --counterparty openai

Why raw tool execution is not enough

Raw OpenClaw or MCP tool execution exposes capability. NORNR adds the missing layer above capability:

  • should this action clear under the active mandate
  • who reviews it when it should queue
  • what finance or audit can inspect afterward

What happens after queued

Queued does not mean "probably yes later". It means:

  1. stop the autonomous action
  2. inspect review-bundle, anomalies or timeline
  3. let a named operator approve or reject
  4. carry the same action into finance-packet, audit-export or monthly-close later

Suggested demo GIF

The clearest first demo is:

  1. the local agent proposes a consequential action
  2. NORNR returns approval_required
  3. the client pauses instead of continuing
  4. an operator approves
  5. the action resumes with one defended record afterward

Operating rule

Do not let OpenClaw proceed with the autonomous action until NORNR returns approved or an operator explicitly approves the queued intent. Treat queued, blocked, anomalous, or prompt-risk posture as operator review states, not autonomous green lights.

More in MCP Tools