# REA: The Tool That Lets AI Reverse Engineer Any App (and How to Use It Right)

> REA is an open-source MCP server that lets your AI agent inspect apps without their source code, with evidence for every conclusion. How it works and how to use it safely.

*2026-10-25T08:11:44Z*

Ever wanted to know how a feature in some app works inside, without having its source code? That is exactly what **REA** (Reverse Engineer Anything) is for. In the first week of October 2026 it became one of the most talked-about projects on GitHub.

Here is what REA does, how it works in five steps, and how I would use it with Claude Code without getting myself into trouble.

## What REA actually is

REA is an open-source **MCP server + CLI** (MIT licence, TypeScript). MCP is the protocol that lets AI agents like Claude Code, Codex and Cursor call external tools. Install REA and your agent gets a new pair of hands: it can inspect a running app, a native binary, an Electron app or a website, and work out how something is built.

The README puts it this way: «One MCP for reverse engineering across binaries, applications, and runtime behavior».

As of 21 October 2026 the GitHub API shows **353K stars** and **88K forks**, and the npm package (`rea-agents`) was downloaded more than 415K times in a single week. These numbers move daily, so check the source before you quote them.

## How it works: 5 steps

1. **The agent:** you ask your agent to understand a feature. The README's own example: «Understand how search works in the Notes app, show me the evidence, and build a similar feature for my project».
2. **REA (MCP):** the agent talks to REA as an MCP server, one more tool it can call.
3. **Analysis engines:** for native apps REA doesn't reinvent the wheel. It drives a reverse-engineering tool you already have, such as **Ghidra**, **IDA** or **Hopper**.
4. **The target:** a native binary, a JavaScript or Electron app, a .NET assembly, a website, and many more formats such as APKs, HAR files and WASM.
5. **An evidence report:** you get a report back where every conclusion carries its evidence and its limits. To me this is the key point: «Analysis runs locally, and results include the evidence and limitations behind each conclusion». It runs on your machine, and the output is more than confident-sounding AI prose.

## How I would use it with Claude Code

I work in the MCP world every day: I build [Orchestra MCP](https://github.com/orchestra-mcp/framework) for agent-driven project management and [Zekra](https://zekra.dev), a shared memory for agents. My plan for REA is simple:

1. **One setup command** from the official source:
   ```bash
   npx rea-agents setup
   ```
   It wires REA into Claude Code, Codex, Cursor, Gemini CLI, Grok Build and others.
2. **A narrow question, not an open one:** "How does search work in this app? Show me the evidence" beats "reverse engineer the whole thing".
3. **Read the evidence before the conclusion:** if the report says a feature uses a certain index, look at the evidence it cites. Weak evidence means a weak conclusion.
4. **Build it my way:** the goal is to understand the idea and build something similar in my own project, not to copy someone else's code.

It's the same principle as in [Multi-Agent Architecture Without the Chaos](/en/blog/multi-agent-architecture-without-chaos): one agent with clear tools beats a crowd of agents.

## Use it right

**1. Official sources only.** The README itself warns about copycats: «A similar repository name or copied README does not establish that a download is official». There are three official sources: [github.com/morluto/rea](https://github.com/morluto/rea), [rea.tools](https://rea.tools) and the [`rea-agents` package on npm](https://www.npmjs.com/package/rea-agents).

**2. Only analyse what you're allowed to.** Per the README, REA is «for lawful research, education, interoperability, software preservation, and analysis of software you own or are authorized to examine». That means your own software, or software you have written permission to analyse, respecting its licence and EULA. Reverse engineering without permission can be illegal even when the tool makes it easy.

**3. Your agent gets more power.** A tool like this runs things on your machine. Use an isolated project and review the commands your agent wants to run.

## The takeaway

REA moves AI agents from "writes code" to "understands existing software", with evidence for each claim. Use it to learn, understand and build better, on software you're allowed to analyse, installed from the official sources only.

Tried it? Tell me what you used it for.

**Sources:** [GitHub: morluto/rea](https://github.com/morluto/rea) · [rea.tools](https://rea.tools) · [npm: rea-agents](https://www.npmjs.com/package/rea-agents) · [GitHub API](https://api.github.com/repos/morluto/rea) (numbers as of 21 October 2026)

---

Source: https://fadymondy.com/en/blog/rea-reverse-engineer-anything-ai-agents