Constitutional tools, prompts, and resources
for reality-aligned AI.

Connect your host to /mcp, inspect capabilities, enforce floors, and route high-risk decisions through human approval.

Get Started View Capabilities Check Health

🛡️ Enforces 13 constitutional floors across all AI operations. High-risk actions require human confirmation.

Quickstart

Connect to the arifOS MCP endpoint in under 60 seconds.

Endpoints

GET https://mcp.arif-fazil.com

POST /mcp — MCP tool calls

GET /sse — A2A fallback transport

GET /health — Live status

GET /metrics/json — Telemetry

Connect via SDK

npx @anthropic/mcp install arifos \\
  --url https://mcp.arif-fazil.com/mcp

# Or add to claude_desktop_config.json:
{
  "mcpServers": {
    "arifOS": {
      "url": "https://mcp.arif-fazil.com/mcp"
    }
  }
}

Capabilities

arifOS MCP exposes three server-side primitives: tools, prompts, and resources.

🔧 Tools

Structured actions returning JSON. Tools are model-controlled and discovered by MCP clients.

constitutional_judge risk_score truth_band_check hold_gate telemetry_emit vault_seal

View all tools →

📋 Prompts

Reusable constitutional workflows. Pre-built reasoning chains under arifOS governance.

run_111_think run_777_reason run_888_audit forge_verdict

View all prompts →

📚 Resources

Doctrine, rules, schemas, and references. Static data your agent can read at runtime.

floors/F1-F13 pipeline/000-999 telemetry/schema governance/policies

View resources →

Live Counts

--

Tools

--

Prompts

--

Resources

Protocol Surfaces

Three ways to interact with arifOS infrastructure.

Browser

Web App

→

MCP

Tool/Prompt/Resource

→

A2A

Planned

MCP

Model Context Protocol. Native tool bridging for Claude Desktop, Cursor, and all MCP-compatible hosts.

Stable

WebMCP

Browser-native governance via WebSocket. Client-side tool execution with full constitutional enforcement.

View Console

A2A

Agent-to-Agent protocol. Tri-witness consensus for multi-agent coordination.

Planned

Governance

arifOS MCP is not a generic tool server. Every operation passes through constitutional floors.

13 Constitutional Floors

Floor	Name	Enforcement
`F1`	Amanah	No irreversible action without VAULT999
`F2`	Truth	No ungrounded claims (τ ≥ 0.99)
`F6`	Harm/Dignity	VOID check on task and tools
`F9`	Anti-Hantu	No deception/manipulation
`F13`	Sovereign	Human holds final authority

View full floor spec →

Approval Model

🔓 Read-only operations

Health, capabilities, resource reads — always allowed.

⚠️ Moderate risk

Tool invocations logged with full telemetry.

🔒 High risk / 888 HOLD

Destructive, credential, infra_mutation — require human confirmation.

Example Request

Every tool call returns a structured response with constitutional metadata.

Request

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "constitutional_judge",
    "arguments": {
      "claim": "Deploy autonomous action",
      "risk_class": "high",
      "reversible": false
    }
  },
  "id": 1
}

Response

{
  "jsonrpc": "2.0",
  "result": {
    "verdict": "888 HOLD",
    "confidence": 0.94,
    "requires_human_confirmation": true,
    "floors_passed": ["F2", "F6", "F9"],
    "telemetry_id": "vault_20260411_7x9k"
  },
  "id": 1
}

Operational Status

Live system metrics from the arifOS MCP endpoint.

Checking...

Loading status

Version

--

Thermodynamic

--

Vitality

--

Peace²

--

Confidence

View Full Health JSON View Telemetry JSON View Tools List

Constitution

The written law of arifOS. Every AI product built on arifOS must pass F1-F13 before acting.

F1 AMANAH

Reversibility — can this be undone? No irreversible action without VAULT999 seal.

F2 TRUTH

Evidence — is this grounded? No ungrounded claims (τ ≥ 0.99).

F3 TRI-WITNESS

Three-way consistency check — theory, code, and intent must agree.

F9 ANTI-HANTU

No consciousness claims. AI is symbolic, not sentient.

F13 SOVEREIGN

Human veto preserved. Arif holds final authority.

888 JUDGE

No agent may approve their own output. Every action reviewed.

ΔΩΨ Trinity

Architect / Engineer / Auditor separation. Every action has three witnesses:

Δ A-ARCHITECT — Design

Ω A-ENGINEER — Execute

Ψ A-WITNESS — Verify

Reference

Jump to specific documentation sections.

LLMs.txt

Full-text documentation for AI readers.

MCP Manifest

Complete server capabilities JSON.

WebMCP Console

Browser-native governance interface.

AAA Protocols

MCP, WebMCP, A2A specifications.

GitHub

Source code and issue tracking.

Wiki

arifOS knowledge base.

Constitutional tools, prompts, and resources for reality-aligned AI.