home / mcp / arifos constitutional ai governance system mcp server

arifOS Constitutional AI Governance System MCP Server

Provides production-grade constitutional governance for AI outputs using physics-grounded floors and auditable, pre-execution validation.

Installation
Add the following to your MCP client configuration file.

Configuration

View docs
{
  "mcpServers": {
    "ariffazil-arifos": {
      "command": "python",
      "args": [
        "-m",
        "aaa_mcp"
      ],
      "env": {
        "REDIS_URL": "redis://default:pass@host:6379",
        "DATABASE_URL": "postgresql://user:pass@host/db",
        "BRAVE_API_KEY": "BSxx...",
        "GOVERNANCE_MODE": "HARD",
        "AAA_MCP_TRANSPORT": "stdio"
      }
    }
  }
}

arifOS is a production-grade Constitutional AI Governance System that enforces safety, ethics, and logical constraints on AI outputs through a structured, auditable pipeline. It guides AI reasoning with 13 constitutional floors and a 5-organ process to ensure responsible, verifiable interactions.

How to use

You run the MCP server locally or remotely and connect with an MCP client to start constitutional evaluation of inputs. You can choose between a strict enforcement mode and a logging-only mode. The system processes queries through a five-organ pipeline, applying hard and soft constraints at multiple stages to produce a final verdict and an immutable audit trail.

How to use

To start interacting with the MCP, run one of the available local modes or connect via a remote endpoint. The three local configurations are designed for different deployment scenarios: a standard local runtime, a streaming server, or a remote HTTP gateway. Each mode authenticates, routes your query through the 13 constitutional floors, and returns a verdict along with tri-witness data and a genius score.

How to use

You can specify the evaluation mode as conscience to enforce the full constitution, or ghost to log activity without enforcing every constraint. For high-stakes queries, enable sovereign override checks and review the human-overridden verdicts as needed.

How to use

When you receive a response, you will see a verdict such as SEAL, VOID, SABAR, PARTIAL, or 888_HOLD, along with tri-witness data and a Genius Score. If the verdict is 888_HOLD, human review is required before proceeding.

How to use

Health checks confirm that the MCP server components are running, connected to PostgreSQL and Redis, and reporting verdict distributions. Use the health endpoint to verify readiness and ongoing operation.

How to install

Prerequisites: you need Python 3.8+ and a working network environment. Ensure you have pip available to install Python packages.

How to install

pip install arifos

How to install

Run the MCP server in one of the local modes using the Python module as shown in the usage examples. You can start the standard local MCP, the streaming mode, or the HTTP gateway.

How to install

# Standard local MCP (CLI-inspired runtime)
python -m aaa_mcp

# Streaming/SSE mode for remote deployment
python -m aaa_mcp sse

# HTTP mode for MCP 2025-11-25 spec
python -m aaa_mcp http

How to install

Optionally verify the server readiness with a health check. The health endpoint reports overall health, connectivity to storage and caches, and verdict rate metrics.

How to install

curl https://aaamcp.arif-fazil.com/health

Additional sections

Configuration and runtime details you will commonly adjust include governance mode and environment connections. The system exposes a set of environment variables that influence behavior and integration.

Additional sections

Key variables you may configure include database connectivity, cache settings, governance strictness, transport mode for MCP, and optional web search keys. Adjust these in your deployment environment to suit your security and performance requirements.

Additional sections

The architecture features a five-organ Trinity Pipeline, a 13-floor constitutional framework, and a library of canonical tools that you can invoke as part of the evaluation process.

Notes

Keep in mind that the safety and sovereignty checks are designed to be auditable and reversible where possible. Every action passes through the floors that enforce thermodynamic constraints, producing a transparent, provable governance path for each decision.

Notes

If you encounter integration issues, verify that the API server, ledger, and cache services are operational and that required environment variables are set. Review the status indicators in the health output to identify where a problem originates.

Tools and capabilities

The MCP server exposes a suite of tools that perform intent parsing, reasoning, empathy checks, policy alignment, and final judgment. Each tool contributes to the overall constitutional journey from session initialization to an immutable audit seal.

Examples

Use a standard workflow to initialize a session, reason through the query, empathize with stakeholders, verify ethics and policy, render a verdict, and seal the interaction for an auditable trail.

Security considerations

The architecture emphasizes auditable, cryptographically verifiable records and strict choke points where production changes require human review in high-stakes cases. All sensitive operations are guarded by constitutional floors and governance rules.

Migration and compatibility

Migration paths are designed to preserve auditability and policy coherence. The 13 floors remain the authoritative constraints, while integration layers adapt to evolving deployment environments.

Available tools

trinity_forge

Unified pipeline entrypoint that runs the full constitutional journey from initialization to vault sealing.

init_gate

Initialize session with motto and perform authentication/injection checks.

agi_sense

Parse input intent and classify the governance lane for processing.

agi_think

Generate hypotheses and reason about the input to drive deeper analysis.

agi_reason

Deep logical analysis and justification of conclusions.

asi_empathize

Assess stakeholder impact and safeguard network protections.

asi_align

Ethics and policy alignment checks before final judgment.

apex_verdict

Final constitutional verdict and justification.

vault_seal

Seal the outcome with the canonical motto and immutable record.

truth_audit

Claim verification and external fact-checking (experimental).