REA is an MCP-based reverse-engineering toolkit that helps agents inspect apps, binaries, and runtime behavior, then explain the evidence behind what they find.
REA is a TypeScript project for reverse engineering across several target types, including native binaries, JavaScript/Electron apps, .NET assemblies, websites, WASM artifacts, EVM bytecode, and recorded Linux crashes. It exposes its capabilities through MCP so agents can investigate targets locally, return evidence-backed findings, and continue into follow-up analysis or implementation work.
It addresses the difficulty of understanding how software works when you do not have the source code or need to verify runtime behavior, binary structure, or implementation details directly. Instead of manually stitching together disassembly, app inspection, and trace evidence, REA gives agents a structured way to gather and explain those clues.
Conceptually, an agent calls REA through MCP to inspect a target, trace relevant code, and receive findings with supporting evidence and stated limitations. The same workflows are also available from the terminal, and setup registers REA with supported agents while adding matching workflow instructions; for some targets it relies on external tools such as Hopper, Ghidra, IDA, pwntools, or WABT, depending on what is being analyzed.
It is attracting attention because it sits at the intersection of AI agents, MCP, and reverse engineering, and it promises a practical workflow for understanding real software behavior rather than only reading source code. The README also emphasizes broad target coverage, local analysis, evidence-backed results, and ready-made setup for popular agent tools such as Claude Code, Codex, Cursor, and Gemini CLI, which makes it feel immediately usable.
The README points to familiar reverse-engineering tools and environments such as Hopper, Ghidra, IDA, pwntools, and WABT as supporting pieces rather than direct substitutes. In practice, the closest alternatives are traditional manual reverse-engineering workflows in those tools, or custom agent setups without REA's MCP-based packaging and evidence-oriented workflow, but the README does not describe a direct one-to-one competitor.
AI-explained · grounded in each repo's README