REA Setup: From Zero to First Command

Step 1 — Run setup (no global install needed)

npx rea-agents@latest setup

REA ships as a single CLI (rea / rea-agents). setup configures integrations for detected AI agents. Useful flags:

• --dry-run — print the resolved plan without applying anything. Run this first if you're cautious about what gets written.
• --client <agent> — target a specific agent (repeatable); --all-detected configures everything found.
• --yes — approve user-owned actions without prompting (add --install-hopper if you also want the Hopper analysis engine).

What setup actually does: it registers an MCP server and a bundled skill with your agents, and records provider configuration. It does not download analysis engines — Ghidra, IDA, jadx and Hopper integrations require you to supply your own installations (see Step 4).

Step 2 — Verify with doctor

npx rea-agents@latest doctor

doctor audits your environment: runtime version, detected clients, and each analysis provider (ghidra / ida / jadx / hopper / kuna). A fresh install shows healthy: false — that's normal. The web/JavaScript commands work with zero optional providers; you only need engines for native binary decompilation.

Step 3 — The Node version gotcha

At the time of writing, rea-agents declares engines: ^22.23.4 || ^24.22.1 || >=26.12.1. If your Node is older (e.g. 22.22.2), npm only warns (EBADENGINE) and the CLI still runs — but doctor flags the runtime as missing_dependency. If you want a clean bill of health, upgrade Node past the declared floor.

Step 4 — Optional: native analysis engines

Only needed for binary/mobile decompilation. Skip this for web and JavaScript work.

• Ghidra — set GHIDRA_INSTALL_DIR to an extracted Ghidra 12.1.x directory.
• IDA — set REA_IDA_MCP_CONFIG to your IDA MCP registration JSON (see REA's docs/ida-provider.md).
• jadx — set REA_JADX_MCP_JAR to an absolute path of jadx-headless-mcp 0.7.1 JAR. REA does not download it for you.
• Hopper — install via rea setup --install-hopper --yes, or set HOPPER_LAUNCHER_PATH.

Step 5 — Your first command

npx rea-agents@latest --help          # 100+ commands, grouped by target
npx rea-agents@latest capabilities    # what providers and side effects exist

Good first real workflows (no engines needed): analyze-web-bundle (static analysis of a page's JS via CDP), analyze-javascript-application (local JS/Electron app), capture-browser-scenario (bounded Playwright scenario). We walk through the first one in full, with real evidence, in the web bundle workflow.

FAQ

Does setup modify my agent configs?

Yes, that's its purpose — it writes MCP registration entries and a skill file for detected agents. Use --dry-run to preview exactly which files it plans to touch before approving.

Can I use REA without any AI agent?

Yes. Everything is available as plain CLI commands. The agent integration is a convenience layer, not a requirement.

Windows / Linux?

REA is a Node CLI, so it runs anywhere Node runs. Our testing is on macOS ARM; native-engine setup steps differ per OS.