REA installs with one command. Plenty of people still fail, and most of them fail in the same place.

Step one: get Node onto a supported version

The README is specific: Node 22.x needs 22.23.4 or later, 24.x needs 24.22.1 or later, and 26.12.1 or newer works. The odd-numbered lines, 23 and 25, are not supported at all.

This is stricter than it looks. If you are on 22.11 or 24.5, the command fails and the error does not necessarily say the version is the problem. The reliable move is to install a 24 LTS through nvm or fnm, switch to it, and confirm both node -v and npm -v respond.

Step two: register it with your agent

Run npx rea-agents setup in a terminal. It scans which agents you have installed, registers REA's MCP server with each, writes matching workflow instructions, and backs up existing configuration before touching it.

Then restart the agent. MCP configuration is read at startup, so skipping this makes the install look like it did nothing.

One requirement gets missed constantly: run this from a directory outside the project you intend to analyse. The README asks for package acquisition in a trusted directory.

If you only want the command line, npm install --global rea-agents, then rea --help for the subcommands.

Step three: decide which engine backs it

This determines what you can actually analyse.

JavaScript, Electron, .NET and a broad set of binary formats need no engine. The static analysers are built in.

Native binaries need a local engine. There are four options. Ghidra is free and open source but wants JDK 21 or newer. Hopper is a paid macOS and Linux tool that is noticeably faster on large files. IDA Pro is the most expensive and the most complete. Kuna is another decompiler option. REA does not decompile; it routes the request to whichever of these you have.

Setup can install Hopper for you, with your approval. Ghidra, IDA and Kuna use existing installations.

On Linux there is one more prerequisite: static JavaScript analysis, Ghidra, JADX and the firmware tools all need a procps-compatible ps on PATH for process ownership checks. Without it, commands end with an error that gives no hint of the cause.

Step four: run the first analysis

The fastest success needs no engine. Point it at an extracted JavaScript or Electron app directory:

npx -y rea-agents@latest analyze-javascript-application /absolute/path/to/app --json

On Windows the path looks like D:/apps/example. The result lists modules, imports, Electron process boundaries, and the evidence behind each one.

If you want a guided first run, the site offers a small Notes sample: trace its CSV export, then change one input and see what moves.

Step five: know how to update

REA ships fast. Versions 6.1 through 6.9 all landed inside a single month. With a global install, rea update handles it, and rea --update is an alias. With npx, re-run npx rea-agents@latest setup and restart the agent.

The four usual failures

Node version, as above. This accounts for most of them.

Not restarting the agent. The config is on disk; the running process never read it.

A client that truncates the tool list. REA exposes 207 tools and some clients cannot handle the complete tools/list frame. There are two switches: set REA_MCP_OUTPUT_SCHEMA_PROFILE=compact to bound the advertised output schemas, or off to omit them. In the other direction, if you need full field guidance and nested validation, set REA_MCP_INPUT_SCHEMA_PROFILE=full and pay for it in context.

The wrong package. Lookalike repositories and download sites copy the README. Only three sources are official: morluto/rea on GitHub, rea.tools and rea-agents on npm.

Do not start with a large target. Take a small program you wrote yourself, run one investigation, and check whether the evidence it returns is something you can verify. That step prevents a lot of later confusion.