The Pulse
REA Connects Coding Agents to Reverse-Engineering Tools
REA connects coding agents to tools for inspecting binaries, apps and websites through MCP and a command-line interface. Its latest project update describes agent-assisted game reconstructions, while the software returns evidence and limita
AI.info Team · Written with AI ·
REA’s showcased DX-Ball reconstruction passed 3,205 original-x86 test cases and reproduced all 63 bytes of a compiled function, according to the project’s GitHub page. The example illustrates what the open-source tool offers: a way for coding agents to inspect software, trace how a feature works and use the findings to build a separate implementation. REA connects those agents to reverse-engineering tools through the Model Context Protocol (MCP) and a command-line interface.
From a question to a code path
The tool is designed for questions about software whose source code is unavailable, such as how a feature in an application behaves. REA can inspect native binaries, JavaScript and Electron applications,.NET assemblies and websites, then return findings with supporting evidence and limitations. An agent can use those results to explain behavior or write and test an implementation adapted to a developer’s project.
For native analysis, REA works with Hopper, Ghidra or IDA; the project says JavaScript and.NET inspection can work without a native analysis engine. It also exposes its workflows through a terminal, so a developer can use the same capabilities without an agent session. The project describes its analysis as local, while noting that results are sent to the model provider used by the agent.
Setup registers the MCP server and installs workflow instructions for supported agents including Claude Code, Codex, Cursor, Gemini CLI and Grok Build. Users review and approve the proposed configuration changes before restarting their agent. The project requires supported Node.js versions for the standard installation, and deep native analysis depends on a separately available analysis provider.
What the reconstruction results show
REA’s DX-Ball example follows a sound-position calculation through a game’s original executable and checks a reconstruction against both test cases and compiled function bytes. The project also presents work tracing Notion’s Electron clipboard bridge and a reconstruction of a calculation in the DOS game TH04. Those examples show how the authors want REA used: not simply to produce an explanation, but to carry evidence into a result that can be tested.
In a blog post dated October 10, N0zoM1z0 describes using agents to reconstruct several Touhou games and argues that checks and accumulated project knowledge let work carry between investigations. The author writes: “I want another person to be able to examine why we chose an implementation, even if an agent did most of the work.” The statement points to a human review problem alongside the technical one: a plausible reconstruction still needs evidence that another person can inspect.
Local analysis does not settle permission
REA’s documentation says the tool does not upload the target application to a hosted analysis service, but it distinguishes that from the agent’s model provider, which receives the tool results and follows its own data policy. That distinction matters when the target contains proprietary or sensitive software. Local processing describes where REA performs its analysis; it does not establish that a user has permission to inspect a particular program.
The project’s disclaimer places responsibility for authorization and legal compliance on the user and says the software is intended for lawful research, analysis and reconstruction. N0zoM1z0’s post also discusses tensions that can arise when agent-assisted reconstruction draws on earlier community work, including credit and project contribution rules. REA can organize evidence and testing, but it cannot decide whether a project maintainer will accept the resulting work or whether a user is authorized to examine the original software.
Evidence is not the same as original source
REA’s documentation says native analysis returns pseudocode and assembly, while the agent uses those findings to write and test a new implementation. The distinction is material: the project does not claim to recover the original source code or automatically clone an application. Its showcase’s byte-for-byte result applies to the specified DX-Ball function, not to arbitrary programs.
For developers, the practical offer is a structured bridge between established analysis tools and coding agents, with evidence attached to findings. Whether that bridge produces a reliable reconstruction depends on the target, the available analysis provider and the checks applied to the output. The user remains responsible for authorization before analysis begins.