REA connects reverse engineering directly to AI agents

REA is an open-source project that plugs full-scale reverse engineering tools directly into your AI agents via the Model Context Protocol (MCP).

If you see a feature you like in an application and want to understand how it works without source code, REA lets agents like Claude Code, Cursor, and Codex do the heavy lifting right on your machine.

What you can analyze

Instead of asking an LLM to guess what a binary might be doing, REA connects your agent directly to real inspection tooling.

Its coverage is exceptionally broad. It handles native binaries, JavaScript and Electron bundles, .NET assemblies, Android APKs, firmware images, EVM bytecode, network captures, and crash reports.

For JavaScript and Electron apps, it extracts modules, IPC channels, and routes statically without needing a heavyweight disassembler. For native code, it integrates with your existing installations of Hopper, Ghidra, IDA, or Kuna to pull pseudocode, assembly, symbols, and cross-references.

Crucially, REA is evidence-driven. When an agent presents a finding, it attaches the specific assembly instructions, string offsets, or call chains backing it up, along with noted limitations.

How to get started

Everything runs locally under your own user permissions, so no proprietary binaries leave your box.

To run it, you need Node.js 22, 24, or 26 and npm. Setup takes one command: npx rea-agents setup. The installer automatically registers the MCP server with supported agents and configures the matching workflow rules. You can also install it globally via npm or Bun to use the rea CLI directly.

For native binaries, you will need an engine like Ghidra or Hopper installed locally, while APK analysis relies on headless JADX and a standard JDK.

Why it matters

If you regularly tear down competitor software or need to inspect undocumented client libraries, REA is worth setting up today. It bridges the gap between deep reversing craft and agentic automation, saving hours of manual pivoting between disassemblers.

What makes REA smart is its refusal to let AI hallucinate. By forcing the agent to fetch concrete disassembly and traces before drawing conclusions, it turns reverse engineering from a specialized manual art into a reliable, evidence-backed agent call.