Short answer
REA (Reverse Engineer Anything) is an open-source, MIT-licensed MCP server that lets AI coding agents such as Claude Code, Codex, Cursor and Gemini CLI drive reverse-engineering tools. It returns pseudocode, assembly and behaviour, not original source code. It runs analysis locally, but results go to your AI model provider. It is not an Anthropic product, and the only official sources are github.com/morluto/rea and the rea-agents npm package: a fake copy distributed a trojan in October 2026.
- MIT - licence: an independent open-source project, not an Anthropic or OpenAI product
- 1 of 3 - native engines needed for deep analysis: Hopper, Ghidra or IDA
- 0 - lines of original source code recovered: it returns pseudocode and assembly
- 172 - reported downloads of the fake rea-tools/rea “Windows exe” before the warning
REA went viral on GitHub in October 2026, and so did the claims around it. Some posts say it rebuilds any program’s source code. Others, this week, say Anthropic “quietly killed” it. Neither is accurate. This guide separates what the project’s own README and issue tracker confirm from what social posts claim, and gives security and engineering leaders a short policy for using it safely.
What REA is
REA is a Model Context Protocol (MCP) server for reverse engineering. Its README describes it as “one MCP for reverse engineering across binaries, applications, and runtime behavior.” Once registered, an AI coding agent can call its tools to inspect a target and explain what it does. The same tools also work from a terminal.
- Native binaries, including Go and Rust, and GPU code objects
- JavaScript and Electron apps, and .NET assemblies
- Android APKs and devices, and Flutter apps
- Websites, saved network captures and packet captures
- Firmware, PyInstaller executables and Apple crash reports
- Model files (GGUF, safetensors), SQLite databases and EVM bytecode
Deep native analysis needs one engine you install yourself: Hopper, Ghidra or IDA. Static JavaScript and .NET inspection needs none. Supported agent clients include Claude Code, Codex, Cursor, Gemini CLI and Grok Build, and any agent that supports local MCP servers can register it by hand.
What REA can and cannot do
ClaimRealityWhy it mattersIt reconstructs the original source codeFalse. It returns pseudocode, assembly and how functions relateOutput needs expert review; it is not a copy of the vendor’s codeIt replaces Ghidra or IDA ProFalse. It drives Ghidra, IDA or Hopper, which you install and licenseCommercial tools still need your own valid licenceIt works on any programPartly. Results depend on the format and the tools installedExpect gaps on packed, obfuscated or unusual targetsEverything stays on your computerAnalysis runs locally, but results go to your AI model providerTreat confidential binaries as data you are sending to that providerIt is an Anthropic productNo. It is an independent MIT project that supports several agentsAnthropic cannot “remove” it; check the repository for its status
Did Anthropic block REA?
Reddit conversations this week claimed Anthropic had “quietly killed REA” after viral demos of people analysing smart-TV firmware. We found no announcement, release note or press report of any such change, and the official repository carries no notice about it. Because REA is an independent project, Anthropic could at most change how Claude responds to particular requests, not remove the tool itself.
If an assistant declines a specific reverse-engineering request, that is a model-level decision about that request. It is worth reporting with an example, but it is not evidence that the tool was taken down.
The fake REA “Windows exe” trojan
On 13 October 2026 a user reported, in issue #2424 on the official repository, that a lookalike repository, rea-tools/rea, had copied the README and added a “Windows Standalone .exe” section. Its release served rea-x64.7z (82.6 MB) containing rea-x64.exe, reported as downloaded 172 times.
The reporter cited Microsoft Defender detections (Trojan:Win32/Wacatac.B!ml for the extracted file) and 8 of 45 engines on VirusTotal flagging it as a trojan or dropper. The maintainer closed the issue the same day with a README warning. REA has no official Windows installer: it is installed through npm.
If anyone downloaded it
Treat any machine that ran rea-x64.exe from rea-tools/rea as compromised: isolate it, run a full scan, and rotate any credentials, tokens and SSH keys used on it. Then reinstall REA from npm.
How to install REA safely
Install only from the official project. The package name is rea-agents and it needs Node.js 22, 24 or 26.
# Check the package before installing: it should point to github.com/morluto/rea
npm view rea-agents repository.url
# Register REA with your agent (Claude Code, Codex, Cursor, Gemini CLI…)
npx rea-agents setup
# Or install the CLI globally, then verify registry signatures
npm install --global rea-agents
npm audit signatures
Never download REA as an .exe, a .7z archive or from a site other than GitHub, npm and the project’s own pages. Search results and “download” repositories are where the fake copy spread.
The legal line
REA’s README limits it to lawful research, education, interoperability, software preservation and analysis of software you own or are authorised to examine. It grants no licence for commercial tools such as IDA Pro, Hopper or JEB, and says cracked copies, stolen keys and bypassing licence restrictions are not authorised. Contract terms and local law still apply to the software you analyse.
A five-point policy for engineering teams
Before anyone on the team uses REA
- Install only from github.com/morluto/rea or the rea-agents npm package, and block the rea-tools/rea lookalike.
- Analyse only software you own or have written permission to examine, and record that permission.
- Remember analysis results go to your AI model provider: keep confidential or customer binaries out unless your contract allows it.
- Run it on a non-production machine or VM, never on servers that hold production credentials.
- Treat its output as a lead for an expert to verify, not as recovered source code or proof.
Application security
Want AI tooling adopted safely across your team?
We review AI coding agents, MCP servers and their data flows, and set the guardrails that let engineers use them without new supply-chain or data-exposure risk.
Sources
- REA official repository and README (morluto/rea, MIT licence)
- Issue #2424: fake repo rea-tools/rea distributes a trojan “Windows exe” (13 October 2026)
- REA, AI reverse engineering at record pace on GitHub: what is true and what is not (Pasquale Pillitteri)
Originally published at phpscientist.com.
Top comments (0)