FastREA ⚡

FastREA is an ultra-lightweight, zero-overhead Model Context Protocol (MCP) server for binary decompilation. Inspired by the core concepts of morluto/rea, it eliminates heavy multi-agent loops and complex dependencies. Instead, it provides your LLM with a single, laser-focused capability: Convert native binaries into clean C source code instantly using headless Ghidra.

🛠️ Architecture Overview

The system consists of only two files:

1. setup.py: A one-time initialization script that automatically scans your environment, detects Ghidra, and registers FastREA across your IDEs and Claude Desktop config files.

2. main.py: The lightweight stateless MCP engine called by the model to orchestrate headless Ghidra execution and stream pure C output directly into the model context.

Related MCP server: Ghidra MCP Server

📦 Dependencies & Installation

FastREA relies on minimal external components. Ensure you have Python 3.10+ and Ghidra installed locally.

Required Python Libraries

To run the MCP engine, you must install the official Model Context Protocol Python SDK. Run the following command in your terminal:

pip install mcp

🚀 Quick Start

1. Provision Environment

Run the automated configuration script. It will automatically scan common paths for Ghidra and inject the FastREA integration directly into Claude Desktop and Cursor configurations:

python setup.py

2. Execution & Usage

Restart your Claude Desktop application or Cursor IDE. You can now prompt your model naturally to handle binary reversing workflows:

"Use fastrea to analyze the binary at /absolute/path/to/target.exe and explain its primary protection routines."

🔄 Under the Hood Process:

1. The model detects your instruction and triggers the fastrea_analyze MCP tool inside main.py.

2. main.py dynamically writes a clean Java decompile script and boots Ghidra in headless mode.

3. Ghidra exports a flat target.exe_decompiled.c source file directly in the executable's directory.

4. FastREA reads the file contents and provides the structured code back to the LLM.

5. The model parses the pristine raw C source code flawlessly without any context bloat.

📄 License

This project is open-source and licensed under the MIT License.