By LaunchVault editorial · Official sources checked
REA (Reverse Engineer Anything) connects AI agents to tools that inspect software, from application behavior to native binaries. REA exposes those capabilities through the Model Context Protocol, or MCP, so an agent can ask focused questions and receive evidence from analysis tools. It also provides terminal commands. This review covers the official documentation, installation workflow and practical trade-offs; we have not independently benchmarked its reconstruction accuracy. The useful starting point is a specific question about software you are authorized to examine, rather than a promise to recreate an entire application automatically.
What is REA?
Think of REA as the connection between an AI assistant and an investigation toolkit. The assistant decides what to inspect, REA calls the appropriate tools, and returned findings help guide the next question. Native analysis can use Hopper, Ghidra or IDA; REA does not replace those engines with its own language model. Other workflows inspect application files without opening a native decompiler.
Its official GitHub repository is morluto/rea, and the repository identifies rea.tools as the project website. Use those sources to find releases and instructions. A similarly named repository is not evidence of an official distribution. GitHub stars indicate attention, not verified compatibility, security or accuracy; the count shown on this page is a dated snapshot.
Sources: REA README
What can REA reverse engineer?
For native binaries, the chosen analysis engine can supply assembly, pseudocode, strings, symbols and references. An investigation might trace a particular calculation or follow a function call. Supported file formats and operating systems depend on the provider, so confirm that your chosen engine handles your target before expecting the agent to explain it.
For JavaScript and Electron applications, REA documents module relationships, imports and cross-layer feature tracing. Its .NET tools inspect assembly metadata and Common Intermediate Language without loading or executing the assembly. These are distinct workflows: a readable method inventory does not mean every native dependency or runtime behavior has been recovered.
Website observation uses a Chrome-family browser and CDP to inspect an existing page, scripts and observed behavior. It cannot reveal a remote service's private source code. Searches involving games or Adobe products do not establish product-specific compatibility or permission. Identify the actual artifact, analysis question and authorized scope instead of treating a famous application name as a supported-target guarantee.
Sources: Native investigation · JavaScript application workflows · Managed-code analysis · Website observation
How to install REA (GitHub / MCP)
Step 1: check the runtime. At this review, the published rea-agents package is version 6.5.0 and specifies Node.js 22.19 or newer within 22.x, 24.11 or newer within 24.x, or 26 and above. The installation guide also lists provider-specific prerequisites. Main-branch documentation can describe changes beyond a published release, so check the release you actually install.
Step 2: run npx rea-agents setup in your terminal. Select the agents you want to configure, read the proposed paths and changes, and approve only the intended integrations. Setup registers the MCP server and matching workflow instructions. Optional Hopper installation is a separate decision; an existing Ghidra or IDA installation needs its corresponding configuration.
Step 3: restart the selected agent and confirm the integration is available. Begin with a small artifact you own and a question whose answer you can check. If setup fails, use the official installation diagnostics rather than repeatedly changing unrelated configuration. Linux users should check the documented procps-compatible ps prerequisite for affected workflows.
Sources: Installation and setup
Using REA with Claude Code & Cursor
The setup flow supports Claude Code, Cursor, Codex and other documented clients. Choose the client you actually use; installing a skill file alone is not the same as confirming that its MCP server and analysis provider work. After restarting the client, ask it to describe which tools are available and what additional dependencies your task needs.
A useful first request is to explain one function in your own sample application and identify the observations supporting the explanation. Keep the target and question narrow enough to compare with known source or behavior. Treat generated explanations as hypotheses until you verify them. Client configuration and tool readiness should be checked again when changing providers or upgrading REA.
Sources: Installation and setup · Agent configuration
Is REA free? License and tool costs
REA-authored code uses the MIT license. That makes the project's code available under those license conditions; it does not bundle a paid AI subscription or commercial reverse-engineering software. Budget separately for your chosen model service, analysis engine, hardware and investigation time. We are not quoting a single all-inclusive monthly price because these components are separate products.
Hopper and IDA access remains governed by their vendors' licenses. REA does not provide activation rights, license keys or permission to exceed trial restrictions. Third-party components retain their own terms. Check both the toolchain and the target software when deciding whether you can publish an analysis or reuse an output in a commercial project.
Sources: MIT license · Responsible use and limitations
Is REA legal & safe to use?
The project describes authorized research, education, interoperability and software preservation as intended uses. Its disclaimer rejects unauthorized access, copying protected code or assets, and bypassing licensing or other protection measures. The tool's availability does not establish permission to analyze every target. Determine the rights and scope applicable to your own project before proceeding.
Local analysis also does not guarantee that an entire AI workflow stays offline. Review your agent's data handling and the actions requested from external tools. Use an appropriate environment for untrusted files, and keep credentials or private captures out of material shared with a model. We have reviewed the documentation, not performed a security audit or verified that every workflow is harmless.
Sources: Responsible use and limitations
REA review: pros & cons
The strongest reason to try REA is workflow continuity: an agent can move between a question, tool observations and a follow-up investigation. MCP and CLI access offer different ways to work with the same toolkit. Evidence-oriented output is useful when you need to inspect why a conclusion was reached rather than accept a fluent answer at face value.
The trade-offs are setup effort, provider dependencies and the need for technical judgment. Decompilation and AI interpretation can be incomplete or wrong, and frequent changes require version awareness. REA looks most relevant to developers and analysts with a concrete investigation; it is a poor fit for someone expecting a one-click, verified replacement for proprietary software.
Sources: REA README · Responsible use and limitations
REA alternatives
Start by deciding which part of the workflow you want to replace. Direct work in Ghidra may suit someone who prefers inspecting a binary manually. Rizin is another reverse-engineering framework worth evaluating for a command-line workflow. These are analysis tools, while REA connects tools to agents; choosing one does not necessarily exclude using it with an agent integration.
The related LaunchVault entries below cover adjacent MCP workflows. MCP Explorer focuses on creating MCP servers, and APIMCP.dev connects REST APIs to MCP. They may help with integration tasks, but neither listing establishes an equivalent native-binary investigation capability. Compare the task you need to perform, not just whether a product mentions MCP.
Sources: Ghidra project · Rizin project