REA is a reverse-engineering toolkit for developers who want their coding agent to inspect software and explain how a feature works.
Ask about a feature, then trace the implementation
REA, short for Reverse Engineer Anything, connects an agent to analysis tools through MCP. The agent can inspect a target, follow related code and return an explanation with supporting evidence. It can then use that understanding to implement and test a similar feature in your own project. The useful starting point is a bounded question about a program, rather than a request to recover an entire application at once.
The official website uses Windows Calculator as an example: why does entering 200 + 10% give 220, while 200 multiplied by 10% gives 20? Its walkthrough traces the percentage handler and explains how the selected operator changes the calculation. That is a developer-supplied demonstration, not an investigation performed by Seailife.
Different targets need different tools
- Native executables and libraries: inspect pseudocode, assembly, strings, symbols, calls and references with a configured Hopper, Ghidra or IDA provider. A provider must be available before this analysis can work.
- JavaScript and Electron applications: inspect extracted files or an ASAR archive for modules, imports, source maps, routes, IPC connections and native add-on relationships. Static inspection does not need a native decompiler.
- Websites: inspect page structure, loaded scripts and network observations through a Chrome-family browser. Browser analysis requires the connection described in REA's guide.
- .NET assemblies: read metadata, CIL instructions and declared native dependencies through static analysis.
- Other artifacts: the repository documents separate workflows for Android APKs, firmware, saved network captures and recorded crashes. These have additional tool and operating-system requirements; they are not all enabled by the same basic install.
Static JavaScript and .NET inspection reads supplied files without running the application. Runtime capture is different: it runs or interacts with the selected target using your permissions. Choose the workflow based on whether the question needs file evidence, live behavior or both.
Evidence is part of the answer
The official repository describes a loop in which the agent asks a question, REA inspects relevant code, and the agent uses the returned evidence to decide what to check next. Results include the limitations behind conclusions. An explanation inferred from a function is different from behavior that has been reproduced against the original program.
- The DX-Ball showcase follows a sound call into a position-to-pan calculation and reconstructs it in C. The authors report comparison against 3,205 original-x86 cases and all 63 compiled function bytes for that specific reconstruction.
- The Notion showcase follows a clipboard API through the Electron renderer, preload layer, IPC and main process to inspect the rich clipboard format.
- The website's dinosaur-game example inspects the running script, recovers its speed and acceleration settings, and demonstrates a small reconstruction with adjustable speed.
These examples show what a focused investigation can produce. They do not establish that every application can be fully reconstructed, or that generated code will match a target without further checks. The product image on this page is the project's own screenshot of a Hopper analysis bridge, not a Seailife test result.
Installation connects the toolkit to your agent
The documented setup command is npx rea-agents setup. It lets you choose agents, review the proposed configuration changes, and register the MCP server with matching workflow instructions. Existing configuration is backed up, and the agent needs to restart afterward. A command-line interface provides analysis workflows without requiring an agent to drive every step.
The repository currently requires Node.js 22.x from 22.19, 24.x from 24.11, or 26 and newer, plus npm. Some specialized operations have stricter requirements. Native provider support varies by host and file format; Windows Ghidra support is documented as experimental. Features recently added to the main branch may also be newer than the latest published npm package.
The toolkit and the analysis costs are separate
REA is released under the MIT license. That covers the toolkit, rather than everything connected to it. Your coding agent's model usage and any commercial analysis provider are separate choices and costs. The setup can optionally install Hopper with approval, or use an existing supported provider.
The project's stated use cases include research, education, interoperability and software preservation, for software you own or are authorized to examine. For a first investigation, select one feature, ask for the source locations and unresolved questions, then compare a small reconstruction with the original behavior before expanding the scope.