REA: Agentic Reverse Engineering Without Source Code

The open-source MCP server REA (morluto/rea, MIT) enables agents in Claude Code and Codex to reverse-engineer software without source code — from native binaries via Ghidra, Hopper, and IDA to .NET assemblies, APKs, firmware, and websites — and has gathered 240k stars on GitHub by mid-October 2026.

Compiled by Sergey KostenchukPublished 2026-10-20Updated 2026-10-20

2026-10-20 Coding Anthropic

🛠 REA: Agentic Reverse Engineering Without Source Code

The open-source MCP server REA (morluto/rea, MIT) allows agents in Claude Code and Codex to analyze software without source code: native binaries via Ghidra, Hopper, and IDA, .NET assemblies, APKs, firmware, and websites. It is installed via npx rea-agents setup and runs locally. It has 240k stars on GitHub.

🌍 REA turns classic disassemblers into tools controlled by an LLM agent via MCP. The key feature is the Evidence Contract: observations, inferences, and unknowns are explicitly separated, addressing the main weakness of LLMs in reverse engineering — hallucinations.

👤 In a test by MG AI Blog, the Claude Code + REA combination correctly analyzed the structure of an iOS screen feature, where the agent without REA had confused the order of cards. However, it is about 3.5 times more expensive and slower, and REA does not restore code — only pseudocode and evidence.

Source 1: https://github.com/morluto/rea

Source 2: https://mg-cafe.github.io/blog/reverse-engineer-anything-claude-code/

Sources