REA, short for "Reverse Engineer Anything", is a free, open-source tool that lets an AI coding agent study how an existing app works. It went viral on GitHub in October 2026. This guide gives you the official repo, the install commands, and starter prompts, plus the legal line you should not cross.
What you need
The repo: github.com/morluto/rea. It is MIT licensed. The README says to get it only from official sources: that repo, rea.tools, or the
rea-agentsnpm package. Similar-named repos exist, so check the name.An AI coding agent. Setup supports Claude Code, Codex, Cursor, Gemini CLI and Grok Build, and any agent that supports local MCP servers. Claude Code comes with Claude Pro, Max, Team or Enterprise, or a Claude Console account.
Node.js and npm. The README asks for Node.js 22.x (22.19 or later), 24.x (24.11 or later) or 26+.
Optional extra tools, depending on what you analyse. Native apps need Hopper, Ghidra or IDA. Android APKs need JADX and a JDK. Firmware needs Binwalk or Unblob. JavaScript, Electron and .NET apps need no extra engine.
Step by step
1. Install Node.js from the official site if you don't have it. Check with node -v in your terminal.
2. Connect REA to your agent. This registers REA with your agent and adds its workflow instructions.
npx rea-agents setupChoose your agent (Claude Code, Cursor and so on), review the config changes it proposes, approve them, then restart your agent.
3. (Optional) Install the command-line tool.
npm install --global rea-agents
rea --help4. Try a quick test on an Electron or JavaScript app you own. Use the full folder path.
npx -y rea-agents@latest analyze-javascript-application /absolute/path/to/app --json5. Ask your agent one narrow question. Pick one feature, not a whole app. Here is the example prompt from the README:
Understand how search works in the Notes app, show me the evidence, and build a similar feature for my project.6. Use these starter prompts. They follow a method popularised by creator Nate Herk: one feature at a time, a short summary first, and every finding labelled by how sure the agent is. This method is his, not part of the README.
Understand one feature:
Use REA to study how [feature, e.g. "the floating camera bubble"] works in [app I own or am allowed to analyse]. The app is built with [Electron / native macOS / Android / website]. Pick the right REA tools for that build.
Give me a short summary first: what it does, how it does it, and what you could not work out.
Label every finding as SEEN (found in the evidence, with a citation), INFERRED (a likely guess) or UNKNOWN. Do not treat INFERRED items as facts.Rebuild the behaviour, not the code:
Using only the SEEN findings from your summary, build my own version of this feature in [my stack]. Write fresh code. Do not copy the original app's code, icons, text or other assets. Then write 5 test cases that check my version behaves the same way, run them, and fix anything that fails.Fix something in your own app:
Use REA to find where [slow startup / this IPC channel / this setting] is handled in my app at [path]. Show me the evidence, explain the cause in plain English, and suggest the smallest fix.7. Keep it updated.
rea updateGood to know
Only analyse software you own or are authorised to examine. REA's disclaimer lists lawful research, education, interoperability, software preservation and security work. It rules out bypassing DRM, licence checks or anti-cheat, copying proprietary code or assets, and getting around access controls. App terms of service and local law can also limit reverse engineering. If you're unsure, don't.
Commercial tools need their own licence. REA does not give you IDA Pro, Hopper or JEB. The README says cracked copies and shared keys are not allowed.
"Runs locally" is only half the story. REA analyses the app on your machine, but the results go to your AI agent, so your model provider's data policy still applies. Don't analyse confidential client apps without permission.
Results depend on the model. Check the SEEN evidence yourself before you build on it.
The viral examples are unconfirmed. We could not find sources for the TV app (30 seconds to 3), AirDrop on Linux, or $50 photo frames turned into dashboards. Verified write-ups mention other projects, such as a rebuilt Loom-style screen recorder.
Star counts move fast. The repo showed about 327,000 stars when we checked. We could not confirm it was the number one trending repo of the week.
Sources
What the MCP server does and where the line is (Developers Digest)
REA five-step method: seen, inferred, unknown (OpenClaw Database)
Found this useful? Follow @kevinsingh.me on Instagram for a new AI tool every day, and subscribe so you don't miss the next one.