REA Claude Code Setup Guide

Before you start REA Claude setup

REA Claude setup connects the open-source Reverse Engineer Anything tools to Claude Code through MCP. This page focuses on installing, checking and diagnosing that connection. For an explanation of what the combination can do, read what Claude REA can help you understand before choosing your first target.

REA Claude prerequisites

Begin REA Claude setup in the same environment where you run Claude Code. Check that your terminal can start Node.js and npm, and confirm that Claude Code itself launches successfully. Resolving a missing agent installation first makes troubleshooting much more specific.

The current REA Claude runtime requirements follow REA's published installation documentation: supported Node.js lines include 22 from 22.19, 24 from 24.11, and 26 or newer. Check the current official requirements when installing later. Our general getting started guide explains the baseline tools.

node --version
npm --version
claude --version

For REA Claude work, keep terminal output from these checks. A version number confirms that a command is available in that terminal; it does not prove the future MCP process has the same environment. That distinction helps diagnosis when a client cannot find an executable.

Decide the REA Claude configuration scope

Use REA Claude setup for the Claude Code profile you actually intend to use. The official client identifier is claude_code. If you use a custom CLAUDE_CONFIG_DIR, keep it consistent between setup and launching the agent. A different profile can make valid REA Claude registration appear missing.

Run REA Claude setup and review the plan

The recommended REA Claude route uses REA's setup command rather than an invented downloadable installer. Each Get REA button on this page opens the homepage setup section. There you can copy the existing installer prompt or command and begin REA Claude setup in your local environment.

Preview the REA Claude changes

If you want to inspect REA Claude configuration before applying it, request a dry run. The following command selects Claude Code explicitly and asks setup for a read-only plan. Review the paths, workflow destination and optional provider requirements before deciding whether to apply the REA Claude changes.

npx -y rea-agents@latest setup --client claude_code --dry-run

For REA Claude setup, npm package-download approval and REA's configuration approval are separate steps. The dry run above does not register a working server. A useful review checks which client is selected, which files would change and whether an optional analysis engine is relevant to the intended task.

Apply the selected REA Claude setup

Run the setup command below when ready, review the proposed plan and approve the intended changes. REA Claude setup normally includes the bundled workflow with the selected client integration. Optional analysis-provider setup depends on your choices and environment; installing every provider is not a requirement for every investigation.

npx rea-agents@latest setup --client claude_code

Keep the completion message, including any failed checks and requested restarts. Do not interpret a cancelled plan as a finished installation. REA Claude setup should report what it configured; the next step is to verify that the restarted client can actually use the registered tools.

Verify the REA Claude connection

Restart the REA Claude client session

After applying REA Claude setup, restart Claude Code as requested by the completion message. A conversation that was already open may not reflect the new registration. Reopen the intended profile and project so your check tests the environment you will use for the investigation.

The standard REA Claude configuration uses Claude Code's personal MCP configuration, with the bundled workflow in its personal skills directory. Custom profile configuration can change the root. Prefer the exact paths reported by setup over copying another machine's paths into your environment.

Inspect REA Claude registration and readiness

Use the following checks for REA Claude diagnosis. The first asks Claude Code about registered MCP servers. The second asks REA for scoped client and workflow diagnostics. Read the reported status and paths; this guide intentionally provides no fabricated success output that could disguise a local failure.

claude mcp list
npx -y rea-agents@latest doctor --client claude_code --skill --json

A REA Claude doctor report checks local readiness; it does not establish a live agent connection by itself. Follow it with a small task in Claude Code that actually uses REA and returns evidence. That final check distinguishes a configuration file on disk from a usable workflow.

Ask for a small REA Claude evidence check

Choose a known sample and ask REA Claude to inspect its format or one relevant module. Require the agent to show which tool result supports its answer. A general explanation written without a tool call is not a completed connection test, even when the explanation sounds plausible.

1. RegistrationThe intended client knows about REA.
2. ReadinessThe required local tools pass their checks.
3. EvidenceA live tool call returns relevant target information.

Your first REA Claude investigation

Once REA Claude is connected, start with the Notes CSV export investigation. Use a temporary example note and export it once before tracing the action. A bounded REA Claude task gives you an observable file to compare against the agent's explanation and avoids an open-ended rebuild request.

A focused REA Claude prompt

Use REA to inspect the Notes example from the first investigation guide.
Trace the CSV export action and show the relevant evidence.
Explain the output path and suggest a changed input to test the rule.
Do not modify the target or build a replacement yet.

This REA Claude prompt requests an evidence trail and a proposed test, not a guaranteed result. The sample's actual tool outputs depend on your environment. Ask REA Claude to describe any missing access or provider requirement before treating its conclusion as an explanation of the inspected application.

Verify one REA Claude prediction

After REA Claude explains an export rule, change a single input and predict what should happen. A comma or newline can test CSV handling; a different destination can test path selection. Save the resulting file and compare it with the prediction before expanding the task.

Keep the REA Claude evidence, explanation and test result together in your own notes. If they disagree, ask a narrower follow-up question. The Claude REA evidence checklist describes useful outputs. REA Claude setup is complete when the tools support your investigation, rather than merely appearing in a menu.

REA Claude troubleshooting by symptom

REA Claude tools do not appear

First check whether REA Claude setup was approved and whether Claude Code was restarted. Then compare the profile used by setup with the profile used to launch the client. A custom configuration directory is a common place to investigate when registration seems absent despite a successful setup report.

Check the REA Claude client list again and retain the diagnostic paths. A workflow file and an MCP server registration serve different purposes. If only the workflow exists, rerun setup for the intended client and review its plan. Avoid deleting unrelated configuration to repair a specific entry.

REA Claude reports an executable or runtime error

Compare the Node and npm commands available to your terminal with the environment in which Claude Code starts the server. REA Claude errors about a missing command require an environment fix, while unsupported runtime versions require an appropriate Node release. Repeating an identical prompt will not resolve either cause.

Capture the first meaningful REA Claude error, including the command and relevant version numbers. Keep private paths or credentials out of public reports. The official installation documentation is the source for supported runtimes and package-runner options; use it before replacing the REA Claude setup command with an unofficial script.

REA Claude connects but cannot analyze a native target

A working REA Claude connection does not imply an installed native analysis engine. Read the readiness result for the provider required by that target. Use the native guide to choose an appropriate path. REA Claude can require different prerequisites for native inspection than for JavaScript inspection.

Also ask REA Claude to identify the actual target file and format. Selecting a launcher instead of the relevant library can send an investigation in the wrong direction. For an Electron application, the JavaScript and Electron guide may be the more appropriate REA Claude workflow.

REA Claude answers without evidence

Ask REA Claude to identify the tool call and exact result behind its claim. If no live call succeeded, return to the connection checks. If inspection succeeded but the result is inconclusive, narrow the question. REA Claude should be allowed to report uncertainty rather than fill gaps with an invented trace.

When a REA Claude test fails, preserve the changed input and actual output. Ask which assumption the mismatch challenges and what additional evidence could resolve it. That turns a failed prediction into a useful next step instead of allowing the agent to redefine success after observing the outcome.

Maintain a reliable REA Claude environment

Before updating REA Claude, record the working package version, client profile and relevant provider versions. Read the current official maintenance instructions, apply the intended update and repeat a small known investigation. This gives your REA Claude environment a practical regression check without assuming that a completed update proves every workflow still works.

Use the same REA Claude checklist after changing computers, terminals or configuration profiles: registration, readiness and a live evidence call. An old conversation summary cannot establish the state of a new machine. Keeping that distinction explicit helps you diagnose REA Claude problems without repeatedly reinstalling tools that are already configured correctly.

Where to continue after REA Claude setup

For REA Claude concepts and expected results, return to the Claude REA overview. For hands-on work, follow the first investigation and then select a target-specific analysis guide. The FAQ provides broader help. These routes complement this REA Claude connection guide rather than requiring another installation.

Ready to begin REA Claude setup? Use the homepage installer section, review the proposed changes and restart the selected agent. Keep the diagnostic commands nearby and finish with a small evidence check. A verified REA Claude connection gives you a dependable starting point for the next question you want to investigate.