Nanosaur2-670M
well9472 · Kit
Makes illustrations from text prompts with a small 670M-parameter diffusion model, run through ComfyUI. Built as a low-budget research model for people who wan…

An MCP server and CLI by morluto that lets your coding agent study binaries, desktop and Android apps, websites and firmware on your own computer, and back each finding with evidence.
REA: Reverse Engineer Anything is an MCP server for AI apps in the Developer category, listed by Mindket. It is free. This is version 1, last updated on . Add it to Claude Code with the line on this page, or put its JSON in Claude Desktop, Cursor or another app. Nothing runs until you add it.
npx rea-agents setup
REA is morluto’s MCP server and command-line tool for reverse engineering. Your coding agent calls it to inspect an app without its source code, trace how a feature works and show the evidence. Every finding comes back with the limits of what REA could see. Analysis runs on your computer.
People who have a program but not its source, and need to know how it works with proof they can check. Most of them work in security or build software. The README also names preservation work and CTF practice among its lawful uses.
| Industry | Who uses it | How they use it | Example |
|---|---|---|---|
| Security | Security researchers, malware and incident analysts | Open a suspicious binary or app, follow what it does and keep the evidence for a report | An incident analyst has the agent find where a sample builds the addresses it contacts, then exports the evidence bundle for the case file. |
| Software development | Developers who saw a feature in another app | Trace how the feature works, then have the agent build their own version and test it | The README’s Notion case follows the clipboard API from the renderer through preload and IPC into the main process. |
| Games and preservation | People who keep old software running | Rebuild a routine from an old binary in a modern language and test it against the original | The README’s DX-Ball case rebuilds the game’s sound-pan routine in C, and the rebuild passes 3,205 original-x86 cases. |
| Education and CTF | Students, teachers and CTF players | Read compiled code as pseudo code, with the agent showing its proof for each claim | A CTF player opens a challenge binary and asks the agent to find the function that checks the flag. |
| Mobile and devices | Android and firmware engineers working on interoperability | Inspect an APK, a phone over adb or a firmware image their own product has to work with | An engineer studies how a vendor’s Android app talks to its device, so their own app can pair with it too. |
If you only work on your own source code, you do not need it.
Any agent that runs local MCP servers can call it.
You can point it at any of these:
One npx line sets it up for the agents you pick.
| Need | Detail |
|---|---|
| Node.js and npm | Node.js 22 (22.23.4 or later), 24 (24.22.1 or later) or 26.12.1 and up |
| A trusted folder | One outside the project you will study, to run setup from |
| A decompiler, for native binaries | Hopper, Ghidra, IDA or Kuna. Setup can install Hopper once you approve it. |
npx rea-agents setup
For the rea command on its own, install it globally:
npm install --global rea-agents rea --help
REA changes quickly, so update often. A global install updates with rea update, and an npx setup with the latest tag:
npx rea-agents@latest setup
Ask your agent in plain words and give the path or address of the target. It picks the tools and keeps the target open across turns. REA has more than 200 tools. This page lists the main ones, at least one for each kind of target.
| Tool | What it does |
|---|---|
| open_binary | Opens a native binary and keeps it open for the next question |
| procedure_pseudo_code | Shows one function as pseudo code you can read |
| search_strings | Finds the text stored in a target, such as messages and paths |
| trace_feature | Follows a feature through the code that makes it work |
| analyze_javascript_application | Maps a JavaScript or Electron app’s modules, imports and Electron boundaries |
| inspect_desktop_application | Looks inside an installed desktop app |
| inspect_managed_artifact | Reads managed code, such as a .NET assembly |
| inspect_android_package | Reads an Android APK |
| inspect_web_page | Reads a website’s page |
| inspect_firmware_regions | Looks at the regions of a firmware image |
| capture_process_scenario | Runs the target and captures what it does, with your user permissions |
| export_evidence_bundle | Packs the findings and their evidence into one bundle |
The terminal runs the same workflows:
npx -y rea-agents@latest analyze-javascript-application /absolute/path/to/app --json
Replace the path with your target, such as "D:/apps/example" on Windows. The result lists modules, imports and Electron boundaries with their evidence.
The README’s own prompt. Swap Notes for your target app and the feature you want.
Understand how search works in the Notes app, show me the evidence, and build a similar feature for my project.
The agent opens the app with REA and follows search through its code. Its answer walks the code paths with evidence for each step, then says what REA could not confirm. Then it builds the feature for your project and tests it.
If you are new. Reverse engineering means working out how a program works when you have the program but not its source. REA gives your coding agent the tools for that and asks it to show proof for each finding. Start with an app you built yourself and ask the agent to explain one feature.
For experts. It puts engines such as Ghidra, IDA, Hopper and JADX behind MCP tools and keeps a target open across turns, so the agent can follow references. Findings carry evidence and stated gaps, which makes a wrong claim easier to catch. Depth depends on the engines you have installed. Objdump and Rizin work too, for offline native inspection, when you supply them.
Four settings in the registration tune it:
| Setting | When to use it |
|---|---|
| REA_MCP_INPUT_SCHEMA_PROFILE=full | Tool schemas are compact by default to save context. This gives the full field guidance. |
| REA_MCP_OUTPUT_SCHEMA_PROFILE=compact or off | Your client caps the tool list, so the output schemas need trimming |
| REA_GHIDRA_STARTUP_TIMEOUT_MS | A large binary needs a longer start in Ghidra |
| REA_HOPPER_STARTUP_TIMEOUT_MS | A large binary needs a longer start in Hopper |
Runs on your computer: your app starts it.
Run this once in a terminal:
claude mcp add rea -- npx -y rea-agents@latest mcp
In Claude Desktop, open Settings → Developer → Edit Config, and add this to claude_desktop_config.json:
{
"mcpServers": {
"rea": {
"command": "npx",
"args": [
"-y",
"rea-agents@latest",
"mcp"
]
}
}
}Add this to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for every project:
{
"mcpServers": {
"rea": {
"command": "npx",
"args": [
"-y",
"rea-agents@latest",
"mcp"
]
}
}
}Add this to .vscode/mcp.json in your project; Copilot in agent mode then has its tools:
{
"servers": {
"rea": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"rea-agents@latest",
"mcp"
]
}
}
}Add this to ~/.codex/config.toml; the Codex CLI and its IDE extension both read it:
[mcp_servers.rea] command = "npx" args = ["-y", "rea-agents@latest", "mcp"]
In Windsurf, open the Cascade panel’s … menu → Open MCP config file, and add this to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"rea": {
"command": "npx",
"args": [
"-y",
"rea-agents@latest",
"mcp"
]
}
}
}Toolsopen_binaryprocedure_pseudo_codesearch_stringstrace_featureanalyze_javascript_applicationinspect_desktop_applicationinspect_managed_artifactinspect_android_packageinspect_web_pageinspect_firmware_regionscapture_process_scenarioexport_evidence_bundle
Listed MCP servers in the same category that share the most GitHub topics with this one.
| Product | Stars | Activity | License | Last change |
|---|---|---|---|---|
| REA: Reverse Engineer Anything This page | 455,418 | Active | MIT | |
| Context7Shares mcp | 64,331 | Active | MIT | |
| Playwright MCPShares mcp | 39,312 | Active | Apache-2.0 |
Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search and on Trendshift. Mindket doesn’t write or check them. Open one to read it where it was posted.
well9472 · Kit
Makes illustrations from text prompts with a small 670M-parameter diffusion model, run through ComfyUI. Built as a low-budget research model for people who wan…
Waqf-AI · Kit
Reads handwritten Arabic text from images and returns it as text. A fine-tuned PaddleOCR-VL model for people digitizing Arabic handwriting.
Wan-AI · Kit
Makes video clips from a text prompt or an image plus text, at up to 720P and 24 frames per second, in the Diffusers format for a single consumer GPU.
heygen-com · Kit
HeyGen’s open-source framework that turns HTML into MP4 video. Its skills teach Claude Code, Codex, Cursor, Gemini CLI and other agents to plan, build, preview…
tt-a1i · Skill
Generates interactive diagrams and explanatory pages from codebases or textual plans. It allows AI agents to visualize workflows, architecture, and complex log…
QwenLM · Kit
Open-source AI coding agent that works in your terminal, IDE plugins, a desktop app, a web page and chat apps, with sub agents, MCP and many model providers. F…