Can someone help me install rea on my computer?

[andrewccosta04-bit]

Need help installing rea on mac

[Sohan-Zakaria]

Here are the simplest ways to get REA up and running on macOS:

1. Prerequisites
   Make sure you have Node.js installed (v22.19+ or v24.11+ recommended). You can check with:
   node -v
npm -v
   (If you don't have Node installed, run brew install node via Homebrew).
2. Install & Run Guided Setup (Recommended)
   You can run the interactive setup wizard directly with npx without needing to pre-install anything globally:
   npx --yes rea-agents@latest setup
   Alternatively, if you prefer having the rea CLI command available globally:
   npm install --global rea-agents
rea setup
   (Or via the install script: curl -fsSL https://raw.githubusercontent.com/morluto/rea/main/install.sh | bash)
3. What Setup Does:

• Agent Integration: It configures REA’s MCP server and workflow skills for whichever agent you use (e.g., Claude Code, Cursor, Windsurf, Claude Desktop, Devin, etc.).
• Analysis Engines: It checks for decompilers/engines (such as Hopper Disassembler or Ghidra). Hopper can be run in free demo mode if you don't have a license.

4. Verify & Next Steps

• Verify the installation:
  npx rea-agents doctor
• Restart your coding agent/IDE so it reloads its MCP configuration and recognizes the new tools.

[Sohan-Zakaria]

@gilgameshxx The two best options with minimal to no refusals for reverse engineering:

1. Qwen 2.5 Coder (32B-Instruct): Great low-level code understanding, strong tool-calling support, and rarely flags decompiled/assembly payloads. Runs great locally via Ollama or via OpenRouter.
2. DeepSeek-Coder-V2 / DeepSeek-V3: High reasoning ability on complex binary logic with very lax safeguard triggers compared to Anthropic or OpenAI.
   If you need a strict zero-refusal guarantee, grab an abliterated weight (like Llama-3.1-70B-Instruct-abliterated or an abliterated Qwen build) and run it through Ollama/vLLM. Those have refusal vectors stripped directly from the model weights.

[Sohan-Zakaria]

@andrewccosta04-bit That depends on your Mac's unified memory (RAM) and whether you want to run it locally or via an API:

1. If running locally on Apple Silicon (M1/M2/M3/M4)
   macOS handles local models efficiently via Metal acceleration in Ollama or LM Studio. Pick based on your RAM:
   ⚬ 16 GB RAM:
   qwen2.5-coder:7b or qwen2.5-coder:14b (Q4). Fast, fits comfortably in memory alongside your IDE and disassembler.
   ⚬ 32 GB – 48 GB RAM (Sweet Spot):
   qwen2.5-coder:32b (Q4_K_M). This is widely considered the best local coding/analysis model for Apple Silicon right now—it has strong low-level comprehension and minimal refusal issues.
   ⚬ 64 GB+ RAM:
   You can step up to quantized 70B models (like an abliterated Llama 3.1 70B) or run the 32B model at higher precision (Q8/FP16).
   Quick Mac install:
   brew install ollama
ollama run qwen2.5-coder:14b # or :32b depending on your RAM

2. If you have an 8 GB Mac, an older Intel Mac, or prefer zero fan noise
   Offload inference to an API so your Mac doesn't throttle or run out of memory:
   ⚬ DeepSeek API (deepseek-coder / deepseek-chat): Inexpensive, fast, and rarely refuses low-level assembly or binary analysis tasks.
   ⚬ OpenRouter: Run Qwen 2.5 Coder 32B or DeepSeek-V3 via an API key with moderation filters disabled.

[BajanBob]

Can you help for windows

[dfcontrolla]

Can you help for windows

I was just looking out for this as well

[Sohan-Zakaria]

@BajanBob Yes, Windows works, just a few differences from macOS. Here’s the Windows version:

1. Open PowerShell
Start menu → type PowerShell → Enter

2. Install Node.js
Check first:

node -v
npm -v

You need Node 22.23.4+, 24.22.1+, or 26+. If it says not recognized or an older version, install Node 24 LTS:

winget install --exact --id OpenJS.NodeJS.LTS

Type Y if asked. Then close PowerShell and open it again, and check node -v again.

3. Run the setup wizard

npx --yes rea-agents@latest setup

If npm asks to download the package, type y.

Windows PowerShell tip: If you get running scripts is disabled on this system, use npx.cmd instead:

npx.cmd rea-agents@latest setup

Pick your coding agent in the list (arrow keys to move, Space to select, Enter to continue), review the plan, and approve it. It will configure REA’s MCP server + skill for that agent and back up your existing config.

You can skip Hopper for the first run — it’s optional.

4. Verify & restart

npx rea-agents doctor

Then fully quit your agent/IDE (not just close the window — quit from the tray icon too if there is one) and reopen it.

Then in your agent, ask:

Check that REA is connected and show me which REA tools you can use.

Tell me which agent you use — Claude Code, Cursor, Windsurf, VS Code, Codex, Claude Desktop? I can give you the exact --client command for it.

[Dragolone]

It's the same flow on Windows. Install a current Node.js LTS first (REA 6.8 needs Node 22.23.4+ or 24.22.1+; check with node -v), then in PowerShell:

npx.cmd rea-agents@latest setup
npx.cmd rea-agents@latest doctor

(npx.cmd rather than npx avoids PowerShell's "running scripts is disabled" error on machines with a restricted execution policy.)

Setup asks which agents to configure. If you use Claude Desktop, quit it fully before approving, otherwise the running app can overwrite the registration. doctor then lists which analysis engines are still missing. On Windows the Ghidra provider is experimental, see docs/windows-ghidra-p0.md.

If it fails with ERR_MODULE_NOT_FOUND under an _npx path, docs/installation.md has a Windows recovery section ("Missing dependencies in an npx installation"). Check your antivirus protection history first, since that's a common cause.

I've sent a few fixes upstream to REA, so if setup errors out, paste the doctor output here and I'll have a look.

[Sohan-Zakaria]

@Dragolone
Thanks so much for the detailed Windows breakdown — this helps a lot!

I didn't know about the npx.cmd trick, I'll use that instead of npx to avoid the PowerShell execution policy error.

Just to confirm my plan:

1. Install Node.js LTS first and check with node -v — need 22.23.4+ or 24.22.1+ for REA 6.8, right?
2. Then in PowerShell:

   npx.cmd rea-agents@latest setup
   npx.cmd rea-agents@latest doctor

3. Quit Claude Desktop fully before approving setup, then restart it after.

I'll try this tonight and paste the doctor output here if anything errors — especially if I hit that ERR_MODULE_NOT_FOUND / _npx issue you mentioned, I'll check my antivirus history first.

Quick question — for Windows, do I even need Ghidra for the first run, or can I skip it and just get the MCP + agent connected first?

Thanks again for sending fixes upstream too 🙏

[Dragolone]

Yep, that plan is right. One small correction to what I wrote earlier: Node 26.12.1+ is accepted too, and the current LTS on nodejs.org is 26.13.0, so the default LTS download works. 22.23.4+ or 24.22.1+ are fine as well.

And no, you don't need Ghidra to get started. Setup registers REA with your agent whether or not an analysis engine is installed. Ghidra only matters once you want to decompile native Windows binaries (.exe/.dll); static JS/Electron analysis, for example, doesn't need any engine.

doctor will list Ghidra as missing and may report healthy: false overall. That's expected and doesn't mean setup is broken. To check just the Claude Desktop registration:

npx.cmd rea-agents@latest doctor --client claude_desktop --json

When you want native analysis later: install Ghidra 12.1+ and a 64-bit full JDK 21+, then point REA at them with GHIDRA_INSTALL_DIR and JAVA_HOME in the REA MCP server's env block. The agent launches REA itself, so setting them only in your PowerShell session isn't enough. There's a full example in docs/windows-ghidra-p0.md, and doctor --provider ghidra --json checks it.

Good luck!

[Sohan-Zakaria]

Perfect, thanks for clarifying all that!

Good to know Node 26.12.1+ is accepted too — I'll just grab the current LTS (26.13.0) from nodejs.org then, so I don't have to worry about picking the right 22/24 version.

And huge relief that I don't need Ghidra to get started. I thought doctor reporting healthy: false meant I broke something, so it's good to know that's expected when Ghidra is missing. I'll use this to check just the Claude Desktop registration instead:

npx.cmd rea-agents@latest doctor --client claude_desktop --json

For now I'll focus on getting the MCP connection working for static JS / Electron stuff — no .exe/.dll decompiling yet. When I'm ready for native analysis later, I'll follow your note about setting GHIDRA_INSTALL_DIR and JAVA_HOME in the REA MCP server's env block (not just in PowerShell), and check it with doctor --provider ghidra --json.

Really appreciate you taking the time to walk through the Windows quirks. Thanks again for the info's 🙏

[evachxji]

On macOS, the current REA 6.8.0 installation flow is:

1. Check Node.js and npm

Open Terminal and run:

node -v
npm -v

REA currently supports:

• Node.js 22.x — 22.23.4 or newer
• Node.js 24.x — 24.22.1 or newer
• Node.js 26.12.1 or newer

Node 23, Node 25, earlier Node 26 releases, and prereleases are unsupported. If your version is missing or outside those ranges, install a supported Node.js release with npm first.

2. Run the guided setup

From a trusted directory outside the project you want to analyze, run:

npx --yes rea-agents@latest setup

Here, --yes only approves npm's package download; it does not approve REA's configuration changes.

Select the agent(s) you use, review the proposed paths and changes, and approve the plan. Hopper is a separate optional choice—you do not need to install it just to connect REA to your agent.

If you use Claude Desktop, quit it completely before approving the setup, because a running instance can overwrite the registration.

3. Verify and restart

npx --yes rea-agents@latest doctor

Then restart the agent(s) selected during setup.

Static JavaScript/Electron analysis does not require a native analysis engine. Deep native-binary analysis may require Hopper, Ghidra, IDA, or Kuna. Consequently, an unscoped doctor report can still complain about an optional engine even when the agent registration itself works.

If setup fails, paste the exact error and doctor output, along with your macOS version, whether the Mac is Apple Silicon or Intel, and which agent you selected.