Use REA with Claude to understand app features and build your own version

The internet turned REA into a story about getting expensive software for free. That is the weakest way to understand it.

The useful version is simpler. You find one feature you like. You ask Claude how it works. REA inspects the software that was actually shipped to your machine and returns evidence. Claude can then explain the behavior, write a specification, and build a separate version for your project.

Read this guide: https://codewithnishant.dev/free/rea-claude-reverse-engineering/ and the official REA documentation linked from it. Help me install REA for Claude Code using npx rea-agents@latest setup. First check my Node.js and npm versions and inspect the current Claude Code configuration without changing it. Use only github.com/morluto/rea, rea.tools and the rea-agents npm package. Show me the setup plan and let REA ask for approval before it writes configuration or installs anything. Do not install unrelated software. After setup, tell me to restart Claude Code, then run npx -y rea-agents@latest doctor --client claude_code --json. Do not analyze any third-party app until I name software I own or am authorized to inspect. Finally, download the official Notes example and use REA to trace Export notes with file and line evidence. Show what was confirmed, what was inferred and what remains unknown. Ask before paid software, external actions or publishing.

01. What REA actually does

Claude asks the questions. REA inspects the target.

REA is a CLI and MCP server for reverse engineering. Claude Code calls its tools to inspect a supported target, follow the code behind a feature, and return evidence with the finding.

That evidence is the useful part. Instead of accepting "this is probably how it works," you can ask for the procedure, module, import, route, IPC message, call graph or network observation behind the answer.

REA describes its model in three stages: decompile, understand and recreate. It does not claim to recover the original source code or clone an entire application automatically.

The goal is not to copy a product. The goal is to understand one behavior well enough to build your version.

02. What REA can inspect

The target decides which extra tool you need

The official README lists the current requirements per target. Start with JavaScript or the official Notes example. You can learn the workflow before adding a native analysis engine.

03. Install REA for Claude Code

Review the plan before anything changes

REA currently requires Node.js 22.x from 22.19, Node.js 24.x from 24.11, or Node.js 26 and newer. Check your environment first:

node -v
npm -v

Then run the official setup command:

npx rea-agents@latest setup

1. Choose Claude Code from the supported agents.
2. Read the exact paths and configuration changes in the plan.
3. Approve the setup inside REA when the plan matches what you want.
4. Restart Claude Code after setup finishes.
5. Run the scoped readiness check:

npx -y rea-agents@latest doctor --client claude_code --json

The setup can register the MCP server and install the matching workflow instructions. Native analysis engines are separate. Static JavaScript inspection does not require Hopper, Ghidra or IDA.

04. Run your first investigation

Use the official Notes app before touching a large target

REA provides a six-file Electron practice app with an Export notes button. Your agent can inspect it without installing or launching Electron.

Paste this into Claude Code:

Download https://rea.tools/examples/notes-example.zip into a new example folder and unpack it. Use REA to find what happens when I click Export notes. Show where notes.csv is written, which fields it contains, and the relevant files and lines.

The official worked answer traces the click through four places:

1. renderer.js handles the button and calls the export bridge.
2. preload.js sends a notes:export message.
3. main.js receives the message and writes the file.
4. csv.js formats the id and title columns.

The result is notes.csv in the user's Downloads folder. More important, each step points back to the file and line that supports it.

05. Pick a useful AI feature

One feature gives you a question you can verify

Do not begin with "clone this whole product." Pick one interaction you use every day and define the behavior you want to understand.

These are investigation ideas, not completed REA findings from me. The answer may live partly in client code and partly on a private server. REA can only show what the available evidence supports.

06. Rebuild without copying code

Evidence becomes a specification before it becomes your product

Use a clean sequence:

1. Investigate: ask one narrow question and collect the evidence.
2. Specify: write SPEC.md with inputs, outputs, states, edge cases and unknowns. Describe behavior, not proprietary source.
3. Build: implement the feature in a separate folder using your own stack, interface and code.
4. Verify: compare your result with the observed behavior and keep the mismatches visible.

Use this prompt after the investigation:

Turn the confirmed behavior into SPEC.md. Include inputs, outputs, state changes, edge cases, evidence references and unresolved questions. Do not copy source code or assets from the inspected app. Build an original version in ./my-version, write tests from the specification, and show which behaviors match.

This is also the safer product habit. Study the behavior. Ship your implementation.

07. The limits that matter

REA is powerful because it keeps the unknowns visible

• Private servers stay private: code and model weights that were never shipped to your machine are outside the target.
• Local analysis still reaches your agent: REA analyzes the target locally, but your coding agent receives the tool results. Your model provider's data policy still applies.
• Decompiled output is a reconstruction: names, types and structure can be incomplete or wrong. Verify important claims.
• Hard targets remain hard: stripped native binaries can require a provider, platform support and deeper manual work.
• REA does not grant licenses: paid analysis tools, app access and protected content still need valid authorization.

The REA project says it is for lawful research, education, interoperability, software preservation and software you own or are authorized to examine. Laws and contracts vary, so this guide is practical guidance, not legal advice.

08. Copyable prompt pack

Keep the question narrow and demand evidence

Install and verify

Install REA for Claude Code with npx rea-agents@latest setup. Show me the setup plan, let REA ask for approval before writing anything, then restart and run the scoped doctor check for Claude Code.

Investigate one feature

Use REA to investigate how [one feature] behaves in [app I own or am authorized to inspect]. Show the function, file, call and evidence behind each conclusion. Separate confirmed facts, inferences and unknowns.

Write the clean specification

Write SPEC.md from the confirmed behavior only. Include inputs, outputs, states, edge cases, evidence references and unresolved questions. Do not include copied proprietary code or assets.

Build and compare

Build an original version from SPEC.md in ./my-version. Use my project's stack. Write tests for the observed behavior, run them, and report every mismatch instead of hiding it.

09. FAQ

The questions to answer before you begin

Does REA upload the app I inspect?

REA analyzes supported targets on your local machine. Your coding agent receives the results, so check the model provider's data policy and your project rules before inspecting confidential software.

Do I need Ghidra, Hopper or IDA to start?

No for the official JavaScript Notes example. Static JavaScript and .NET inspection work without a native engine. Deep native binary analysis needs a supported provider.

Can REA reveal Claude's private model or server code?

No. It can inspect artifacts and behavior available to your machine. It cannot inspect code or model weights that were never shipped to you.

Can I use REA to unlock paid features?

No. Do not use it to bypass access controls. Use the evidence to understand behavior and build an original version you are allowed to ship.

10. Official sources

Product details checked 16 October 2026

• Official REA repository, for the product scope, supported targets, quick start, local-analysis boundary and official download warning.
• Installation and setup, for setup approval, agent registration, readiness checks and provider requirements.
• Your first REA investigation, for the official Notes example and the expected export trace.
• REA responsible-use disclaimer, for authorization and lawful-use guidance.
• REA MIT license.

I have not run REA against Cursor, v0 or Perplexity for this article. Those examples are questions worth investigating, not verified claims about their private implementation.