This GitHub repo took over this month - Reverse Engineer Anything (REA)

One agent, one front door, and receipts for every answer.

• How to answer “what does this app actually do?” with code locations and stated unknowns, not a chatbot’s best guess.

• Why most business software (Electron, web, .NET) needs no disassembler at all.

• The exact three-piece stack, with real costs, and what it deliberately leaves out.

• How the pieces hand off so you never paste decompiler output into a chat window again.

• A verification habit that makes any agent answer defensible in a security review.

• The authorization and isolation guardrails that keep this legal and your laptop safe.

1. Executive Summary

2. Stack Snapshot

3. The Challenge and the Outcome

4. Stack Overview and Architecture

5. Tool Breakdown

6. Alternative Setups

7. Replace / Keep / Add

8. Power User Configuration

9. Common Mistakes

10. Roadmap

11. Operator Principle

12. If I Were Building This Today

• REA is a free, MIT-licensed MCP server that gives your existing coding agent one front door to reverse-engineering tools. It runs the analysis locally and returns the evidence and limitations behind each conclusion. Github

• The leverage isn’t decompilation, which has been free since Ghidra’s 2019 release. The leverage is removing you as the copy-paste layer between the tool and the reasoning.

• Cost is roughly $0 extra if you already pay $20/month for Claude Pro or ChatGPT Plus.

Ask one agent a bounded question, let REA fetch the evidence, and accept only answers that show their work.

Stack Name: The Evidence-First Investigation Stack

Primary Outcome: Verified answers about how software you’re authorized to examine behaves

Difficulty Level: Intermediate. You need to be comfortable in a terminal; no reverse-engineering background is needed for Electron or .NET targets.

Setup Time: About 15 minutes for the no-engine path. Add 30 to 60 minutes for a native engine. Both are estimates.

Monthly Cost Estimate: $0 incremental if you already have Claude Pro at $20/month, which includes Claude Code. Native analysis on a Mac adds a one-time $99 Hopper personal license.

Productivity Impact: Estimate. A bounded question about an Electron, .NET or web target moves from days of forum-digging (or a specialist engagement) to roughly an afternoon. Native binaries take longer and need human review. This is based on the project’s worked examples, not a benchmark.