Reversecore_MCP

🇰🇷 한국어 (Korean)

An enterprise-grade MCP (Model Context Protocol) server for AI-powered reverse engineering. Enables AI agents to perform comprehensive binary analysis through natural language commands.

📋 Prerequisites

Ghidra (Required for Decompilation)

Ghidra is required for advanced decompilation features. The installation scripts automatically install Ghidra to <project>/Tools directory.

Option 1: Automatic Installation (Recommended)

# Windows (PowerShell)
.\scripts\install-ghidra.ps1

# With custom version/path (optional)
.\scripts\install-ghidra.ps1 -Version "11.4.3" -InstallDir "C:\CustomPath"

# Linux/macOS
chmod +x ./scripts/install-ghidra.sh
./scripts/install-ghidra.sh

# With custom version/path (optional)
./scripts/install-ghidra.sh -v 11.4.3 -d /custom/path

What the scripts do:

• Downloads Ghidra 11.4.3 from GitHub (~400MB)
• Extracts to <project>/Tools/ghidra_11.4.3_PUBLIC_YYYYMMDD
• Sets GHIDRA_INSTALL_DIR environment variable
• Updates project .env file

Option 2: Manual Installation

1. Download: Ghidra 11.4.3
2. Extract to <project>/Tools/ or any directory
3. Set environment variable:

   # Linux/macOS (~/.bashrc or ~/.zshrc)
   export GHIDRA_INSTALL_DIR=/path/to/ghidra_11.4.3_PUBLIC_YYYYMMDD
   
   # Windows (PowerShell - permanent)
   [Environment]::SetEnvironmentVariable("GHIDRA_INSTALL_DIR", "C:\path\to\ghidra", "User")
   

   Or add to .env file (copy from .env.example)

⚠️ Note: JDK 17+ is required for Ghidra. Download from Adoptium if needed.

🚀 Quick Start

Docker (Recommended)

# Auto-detect architecture (Intel/AMD or Apple Silicon)
./scripts/run-docker.sh

# Or manually:
# Intel/AMD
docker compose --profile x86 up -d

# Apple Silicon (M1/M2/M3/M4)
docker compose --profile arm64 up -d

MCP Client Configuration (Cursor AI)

Step 1: Build Docker Image

The unified Dockerfile automatically detects your system architecture:

# Automatic architecture detection (works for all platforms)
docker build -t reversecore-mcp:latest .

# Or use the convenience script
./scripts/run-docker.sh

Step 2: Configure MCP Client

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "reversecore": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-v", "/Users/YOUR_USERNAME/Reversecore_Workspace:/app/workspace",
        "-e", "REVERSECORE_WORKSPACE=/app/workspace",
        "-e", "MCP_TRANSPORT=stdio",
        "reversecore-mcp:latest"
      ]
    }
  }
}

{
  "mcpServers": {
    "reversecore": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-v", "/path/to/workspace:/app/workspace",
        "-e", "REVERSECORE_WORKSPACE=/app/workspace",
        "-e", "MCP_TRANSPORT=stdio",
        "reversecore-mcp:latest"
      ]
    }
  }
}

{
  "mcpServers": {
    "reversecore": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-v", "C:/Reversecore_Workspace:/app/workspace",
        "-e", "REVERSECORE_WORKSPACE=/app/workspace",
        "-e", "MCP_TRANSPORT=stdio",
        "reversecore-mcp:latest"
      ]
    }
  }
}

⚠️ IMPORTANT: File Path Usage in Docker

The MCP server runs inside a Docker container. When using analysis tools, use only the filename, not the full local path.

❌ Wrong	✅ Correct
run_file("/Users/john/Reversecore_Workspace/sample.exe")	run_file("sample.exe")

Why? Your local path (e.g., /Users/.../Reversecore_Workspace/) is mounted to /app/workspace/ inside the container. Tools automatically look for files in the workspace directory.

Tip: Use list_workspace() to see all available files in your workspace.

✨ Key Features

🔍 Static Analysis

Comprehensive file analysis and metadata extraction:

• File Type Detection: Identify binary format, architecture, and compiler information (run_file)
• String Extraction: Extract ASCII/Unicode strings with configurable limits (run_strings)
• Firmware Analysis: Deep scan for embedded files and signatures (run_binwalk)
• Binary Parsing: Parse PE/ELF/Mach-O headers and sections with LIEF (parse_binary_with_lief)

⚙️ Disassembly & Decompilation

Multi-architecture binary analysis with intelligent tooling:

• Radare2 Integration: Full r2 command access with connection pooling (run_radare2, Radare2_disassemble)
• Ghidra Decompilation: Enterprise-grade decompilation with 16GB JVM heap (smart_decompile, get_pseudo_code)
• Multi-Architecture Support: x86, x86-64, ARM, ARM64, MIPS, PowerPC via Capstone (disassemble_with_capstone)
• Smart Fallback: Automatic Ghidra-first, r2-fallback strategy for best results

🧬 Advanced Analysis

Deep code analysis and behavior understanding:

• Cross-Reference Analysis: Track function calls, data references, and control flow (analyze_xrefs)
• Structure Recovery: Infer data structures from pointer arithmetic and memory access patterns (recover_structures)
• Emulation: ESIL-based code emulation for dynamic behavior analysis (emulate_machine_code)
• Binary Comparison: Diff binaries and match library functions (diff_binaries, match_libraries)

🦠 Malware Analysis & Defense

Specialized tools for threat detection and mitigation:

• Dormant Threat Detection: Find hidden backdoors, orphan functions, and logic bombs (dormant_detector)
• IOC Extraction: Automatically extract IPs, URLs, domains, emails, hashes, and crypto addresses (extract_iocs)
• YARA Scanning: Pattern-based malware detection with custom rules (run_yara)
• Adaptive Vaccine: Generate defensive measures (YARA rules, binary patches, NOP injection) (adaptive_vaccine)
• Vulnerability Hunter: Detect dangerous API patterns and exploit paths (vulnerability_hunter)

📊 Server Health & Monitoring

Built-in observability tools for enterprise environments:

• Health Check: Monitor uptime, memory usage, and operational status (get_server_health)
• Performance Metrics: Track tool execution times, error rates, and call counts (get_tool_metrics)
• Auto-Recovery: Automatic retry mechanism with exponential backoff for transient failures

🖥️ Web Dashboard (NEW)

Visual interface for binary analysis without LLM:

# Start server in HTTP mode
MCP_TRANSPORT=http MCP_API_KEY=your-secret-key python server.py

# Access dashboard
open http://localhost:8000/dashboard/

Features:

• Overview: File list with upload stats
• Analysis: Functions list, disassembly viewer
• IOCs: Extracted URLs, IPs, emails, strings

Security:

• XSS protection with HTML sanitization
• Path traversal prevention
• API key authentication (optional)

📝 Report Generation (v3.1)

Professional malware analysis report generation with accurate timestamps:

• One-Shot Submission: Generate standardized JSON reports with a single command (generate_malware_submission)
• Session Tracking: Start/end analysis sessions with automatic duration calculation (start_analysis_session, end_analysis_session)
• IOC Collection: Collect and organize indicators during analysis (add_session_ioc)
• MITRE ATT&CK Mapping: Document techniques with proper framework references (add_session_mitre)
• Email Delivery: Send reports directly to security teams with SMTP support (send_report_email)
• Multiple Templates: Full analysis, quick triage, IOC summary, executive brief

# Example 1: One-Shot JSON Submission
generate_malware_submission(
    file_path="wannacry.exe",
    analyst_name="Hunter",
    tags="ransomware,critical"
)

# Example 2: Interactive Session Workflow
get_system_time()
start_analysis_session(sample_path="malware.exe")
add_session_ioc("ips", "192.168.1.100")
add_session_mitre("T1059.001", "PowerShell", "Execution")
end_analysis_session(summary="Ransomware detected")
create_analysis_report(template_type="full_analysis")
send_report_email(to="security-team@company.com")

⚡ Performance & Reliability (v3.1)

• Resource Management:
  • Zombie Killer: Guaranteed subprocess termination with try...finally blocks
  • Memory Guard: Strict 2MB limit on strings output to prevent OOM
  • Crash Isolation: LIEF parser runs in isolated process to handle segfaults safely
• Optimizations:
  • Dynamic Timeout: Auto-scales with file size (base + 2s/MB, max +600s)
  • Ghidra JVM: 16GB heap for modern systems (24-32GB RAM)
  • Sink-Aware Pruning: 39 dangerous sink APIs for intelligent path prioritization
  • Trace Depth Optimization: Reduced from 3 to 2 for faster execution path analysis
• Infrastructure:
  • Stateless Reports: Timezone-aware reporting without global state mutation
  • Robust Retries: Decorators now correctly propagate exceptions for auto-recovery
  • Config-Driven: Validation limits synchronized with central configuration

🛠️ Core Tools

📊 Analysis Workflow

📥 Upload → 🔍 Triage → 🔗 X-Refs → 🏗️ Structures → 📝 Decompile → 🛡️ Defense

Use built-in prompts for guided analysis:

• full_analysis_mode - Comprehensive malware analysis with 6-phase expert reasoning and evidence classification
• basic_analysis_mode - Quick triage for fast initial assessment
• game_analysis_mode - Game client analysis with cheat detection guidance
• firmware_analysis_mode - IoT/Firmware security analysis with embedded system focus
• report_generation_mode - Professional report generation workflow with MITRE ATT&CK mapping

💡 AI Reasoning Enhancement: Analysis prompts use expert persona priming, Chain-of-Thought checkpoints, structured reasoning phases, and evidence classification (OBSERVED/INFERRED/POSSIBLE) to maximize AI analysis capabilities and ensure thorough documentation.

🏗️ Architecture

reversecore_mcp/
├── core/                           # Infrastructure & Services
│   ├── config.py                   # Configuration management
│   ├── ghidra.py, ghidra_manager.py, ghidra_helper.py  # Ghidra integration (16GB JVM)
│   ├── r2_helpers.py, r2_pool.py   # Radare2 connection pooling
│   ├── security.py                 # Path validation & input sanitization
│   ├── result.py                   # ToolSuccess/ToolError response models
│   ├── metrics.py                  # Tool execution metrics
│   ├── report_generator.py         # Report generation service
│   ├── plugin.py                   # Plugin interface for extensibility
│   ├── decorators.py               # @log_execution, @track_metrics
│   ├── error_handling.py           # @handle_tool_errors decorator
│   ├── logging_config.py           # Structured logging setup
│   ├── memory.py                   # AI memory store (async SQLite)
│   ├── mitre_mapper.py             # MITRE ATT&CK framework mapping
│   ├── resource_manager.py         # Subprocess lifecycle management
│   └── validators.py               # Input validation
│
├── tools/                          # MCP Tool Implementations
│   ├── analysis/                   # Basic analysis tools
│   │   ├── static_analysis.py      # file, strings, binwalk
│   │   ├── lief_tools.py           # PE/ELF/Mach-O parsing
│   │   ├── diff_tools.py           # Binary comparison
│   │   └── signature_tools.py      # YARA scanning
│   │
│   ├── radare2/                    # Radare2 integration
│   │   ├── r2_analysis.py          # Core r2 analysis
│   │   ├── radare2_mcp_tools.py    # Advanced r2 tools (CFG, ESIL)
│   │   ├── r2_session.py           # Session management
│   │   └── r2_pool.py              # Connection pooling
│   │
│   ├── ghidra/                     # Ghidra decompilation
│   │   ├── decompilation.py        # smart_decompile, pseudo-code
│   │   └── ghidra_tools.py         # Structure/Enum management
│   │
│   ├── malware/                    # Malware analysis & defense
│   │   ├── dormant_detector.py     # Hidden threat detection
│   │   ├── adaptive_vaccine.py     # Defense generation
│   │   ├── vulnerability_hunter.py # Vulnerability detection
│   │   ├── ioc_tools.py            # IOC extraction
│   │   └── yara_tools.py           # YARA rule management
│   │
│   ├── common/                     # Cross-cutting concerns
│   │   ├── file_operations.py      # Workspace file management
│   │   ├── server_tools.py         # Health checks, metrics
│   │   └── memory_tools.py         # AI memory operations
│   │
│   └── report/                     # Report generation (v3.1)
│       ├── report_tools.py         # Core report engine
│       ├── report_mcp_tools.py     # MCP tool registration
│       ├── session.py              # Analysis session tracking
│       └── email.py                # SMTP integration
│
├── prompts.py                      # AI reasoning prompts (5 modes)
├── resources.py                    # Dynamic MCP resources (reversecore:// URIs)
└── server.py                       # FastMCP server initialization & HTTP setup

🐳 Docker Deployment

Multi-Architecture Support

The unified Dockerfile automatically detects your system architecture:

Run Commands

# Using convenience script (auto-detects architecture)
./scripts/run-docker.sh              # Start
./scripts/run-docker.sh stop         # Stop
./scripts/run-docker.sh logs         # View logs
./scripts/run-docker.sh shell        # Shell access

# Manual Docker build (works for all architectures)
docker build -t reversecore-mcp:latest .

# Or using Docker Compose
docker compose up -d

Environment Variables

🔒 Security

• No shell injection: All subprocess calls use list arguments
• Path validation: Workspace-restricted file access
• Input sanitization: All parameters validated
• Rate limiting: Configurable request limits (HTTP mode)

🧪 Development

# Install dependencies
pip install -r requirements-dev.txt

# Run tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=reversecore_mcp --cov-fail-under=72

# Code quality
ruff check reversecore_mcp/
black reversecore_mcp/

Test Status

• ✅ 852 tests passed
• 📊 76% coverage
• ⏱️ ~14 seconds execution time

📚 API Reference

Tool Response Format

All tools return structured `ToolResult`:

{
  "status": "success",
  "data": "...",
  "metadata": { "bytes_read": 1024 }
}

{
  "status": "error",
  "error_code": "VALIDATION_ERROR",
  "message": "File not found",
  "hint": "Check file path"
}

Common Error Codes

💻 System Requirements

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make changes with tests
4. Run `pytest` and `ruff check`
5. Submit a pull request

📄 License

MIT License - see LICENSE for details.

🔗 Links

• GitHub Repository
• FastMCP Documentation
• MCP Protocol Specification