Superego MCP

Intelligent tool-call review system for AI agents

Overview

Superego MCP Server provides a configurable review system to AI agents to reduce the amount of manual approvals needed as well as provide automated guardrails against dangerous operations. It analyzes incoming tool calls against a set of rules and if no rule is matched it defers to another agent for review or escalation to a human.

Features

• Rule-based interception: Define flexible rules using YAML configuration with advanced pattern matching (regex, glob, JSONPath)
• Multiple actions: Allow, block, or require approval (sampling) based on configurable policies
• Claude Code Hooks Integration: Direct integration with Claude Code for real-time security evaluation
• Multi-transport Support: STDIO, HTTP, and SSE transports for flexible deployment
• Hot reload: Configuration changes are applied without restart
• AI-powered evaluation: Optional AI inference for complex security decisions
• Performance optimized: Request batching, caching, and connection pooling
• Comprehensive monitoring: Built-in metrics, health checks, and performance dashboard
• Structured logging: Comprehensive logging with structured output
• MCP compatibility: Full Model Context Protocol support with FastMCP framework

Quick Start

Installation

# Install with uv (recommended)
uv pip install superego-mcp

# Or install from source
git clone https://github.com/toolprint/superego-mcp
cd superego-mcp
uv sync

Basic Usage

1. Run security evaluation (for Claude Code hooks):

   echo '{"tool_name": "bash", "tool_input": {"command": "ls"}}' | superego advise

2. Start the MCP server:

   # Default STDIO transport
   superego mcp
   
   # HTTP transport on custom port
   superego mcp -t http -p 9000
   
   # With custom config
   superego mcp -c ~/.toolprint/superego/config.yaml

3. Run interactive demo:

   just demo-fastagent-simple

Claude Code Integration

Superego provides seamless integration with Claude Code through hooks:

Setup Claude Code Hooks

# Add hooks for specific tools (recommended)
superego hooks add --matcher "Bash|Write|Edit|MultiEdit"

# Add universal hook for all tools
superego hooks add --matcher "*"

# Use centralized server mode
superego hooks add --matcher "*" --url http://localhost:8000

For complete hook setup instructions and examples, see: Claude Code Hooks Setup Guide

Quick Hook Configuration

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "superego advise",
            "timeout": 5000
          }
        ]
      }
    ]
  }
}

Configuration

Server Configuration (config/server.yaml)

# Server settings
host: "localhost"
port: 8000
debug: false
log_level: "INFO"

# Rule engine settings
rules_file: "config/rules.yaml"
hot_reload: true

# Multi-transport configuration
transport:
  stdio:
    enabled: true
  http:
    enabled: true
    host: "0.0.0.0"
    port: 8000
  sse:
    enabled: true
    port: 8002

# AI inference configuration
inference:
  timeout_seconds: 30
  provider_preference:
    - "claude_cli"
    - "mcp_sampling"
  cli_providers:
    - name: "claude_cli"
      enabled: true
      type: "claude"
      command: "claude"
      model: "claude-sonnet-4-20250514"

Security Rules (config/rules.yaml)

rules:
  # Block dangerous commands
  - id: "block_destructive_commands"
    priority: 1
    conditions:
      tool_name:
        type: "regex"
        pattern: "^(rm|delete|remove|destroy).*"
    action: "deny"
    reason: "Destructive command pattern detected"
    
  # Protect system directories
  - id: "protect_system_files"
    priority: 2
    conditions:
      parameters:
        path:
          type: "glob"
          pattern: "/etc/**"
    action: "deny"
    reason: "System directory access denied"
    
  # Require approval for file operations
  - id: "sample_file_operations"
    priority: 10
    conditions:
      AND:
        - tool_name: ["edit", "write", "delete"]
        - parameters:
            path:
              type: "regex"
              pattern: "^(?!/tmp/).*$"
    action: "sample"
    reason: "File operation requires AI evaluation"
    sampling_guidance: "Evaluate if this file operation is safe"

Development

Setup

# Setup development environment
just dev

# Run tests
just test

# Run with coverage
just test-cov

# Lint and format
just lint
just format

# Type check
just typecheck

# Run all quality checks
just check

Project Structure

src/superego_mcp/
├── __init__.py              # Package initialization
├── cli.py                   # Unified CLI interface
├── cli_eval.py              # Evaluation mode implementation  
├── cli_hooks.py             # Claude Code hooks management
├── main.py                  # MCP server entry point
├── main_optimized.py        # Performance-optimized server
├── stdio_main.py            # STDIO transport handler
├── domain/                  # Business logic and models
│   ├── models.py            # Core domain models
│   ├── pattern_engine.py    # Pattern matching engine
│   ├── security_policy.py   # Security evaluation engine
│   ├── services.py          # Domain services
│   ├── repositories.py      # Domain repositories
│   ├── claude_code_models.py # Claude Code hook models
│   └── hook_integration.py  # Hook integration service
├── infrastructure/          # External services and adapters
│   ├── config.py            # Configuration management
│   ├── config_watcher.py    # Hot reload implementation
│   ├── ai_service.py        # AI inference service
│   ├── inference.py         # Extensible inference system
│   ├── circuit_breaker.py   # Circuit breaker pattern
│   ├── metrics.py           # Prometheus metrics
│   ├── performance.py       # Performance optimization
│   └── logging_config.py    # Structured logging setup
└── presentation/            # API and transport layers
    ├── mcp_server.py        # FastMCP server implementation
    ├── http_transport.py    # HTTP/WebSocket transport
    ├── sse_transport.py     # Server-sent events transport
    ├── handlers.py          # Request handlers
    ├── monitoring.py        # Monitoring dashboard
    └── server.py            # Transport server orchestration

Testing

# Run specific test file
just test-file tests/test_security_policy.py

# Run integration tests
uv run pytest tests/test_mcp_server_integration.py -v

# Run performance tests
just test-performance

# Run load tests
just load-test

Performance Optimization

# Run optimized server
just run-optimized

# Run performance demo
just demo-performance

# Benchmark rule evaluation
just benchmark-rules

API Documentation

CLI Commands

• superego advise - One-off security evaluation for Claude Code hooks
• superego mcp - Launch the FastMCP server
• superego hooks - Manage Claude Code hook configurations

Tool Request Format

{
  "tool_name": "string",
  "tool_input": {
    "parameter1": "value1",
    "parameter2": "value2"
  },
  "session_id": "string",
  "transcript_path": "string",
  "cwd": "string",
  "hook_event_name": "PreToolUse"
}

Security Decision Response

{
  "decision": "allow|deny|sample",
  "confidence": 0.95,
  "reasoning": "Explanation of the decision",
  "risk_factors": ["risk1", "risk2"],
  "matched_rules": ["rule_id1", "rule_id2"]
}

Monitoring

Access the monitoring dashboard at http://localhost:9090/dashboard when running with metrics enabled.

Metrics available:

• Request volume by tool type
• Decision distribution (allow/deny/sample)
• Processing times
• AI inference latency
• Error rates

Troubleshooting

Common Issues

1. Import errors: Ensure proper Python path setup

   export PYTHONPATH=$PYTHONPATH:$(pwd)/src

2. Hook timeouts: Check Superego service availability

   superego mcp --debug

3. AI inference failures: Verify API keys are set

   export ANTHROPIC_API_KEY=your-key-here

Debug Mode

Enable debug logging:

superego mcp --debug

Logs Location

• Server logs: stderr (structured JSON format)
• Hook operations: /tmp/superego_hook.log
• Metrics: http://localhost:9090/metrics

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'feat: add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Development Guidelines

• Follow conventional commits format
• Ensure all tests pass (just check)
• Add tests for new features
• Update documentation as needed
• Maintain type safety with mypy

License

MIT License - see LICENSE file for details

Container Deployment

Docker Quickstart

1. Pull the latest image:

docker pull toolprint/superego-mcp:latest

2. Run with Docker Compose:

# Start in development mode
docker-compose up -d

# Start in production mode
docker-compose -f docker-compose.prod.yml up -d

Container Management

• Comprehensive container usage guide: Container Usage Docs
• Deployment guide: Deployment Guide
• Troubleshooting: Troubleshooting Guide

Links

• Repository
• Issues
• PyPI Package
• Documentation
• Container Documentation