morluto
REA gives an AI agent local inspection tools and returns findings with evidence and stated limitations, rather than unsupported conclusions. Point it at an application, binary, website, capture, or device data and ask it to trace a feature or inspect its structure. Deep native analysis requires a separately installed provider such as Ghidra, Hopper, or IDA.
For Developers with AI coding agents recreating behavior without source code, and automation authors scripting inspections
REA shares reverse-engineering workflows between a terminal CLI and an MCP server, preserving evidence provenance, uncertainty, and session semantics across both adapters. Its layers keep domain and contract semantics separate from provider adapters and application workflows.
AGENTS.md (Product and Authority; Architecture and Cleanup); README.md (How REA works)
MCP server
Creates the evidence-oriented MCP server and optional target session tools.
src/server/createServer.ts
Reverse-engineering workflow
Runs caller-selected provider commands with timeouts, output limits, cleanup, and evidence.
src/application/reverse/ReverseEngineeringService.ts
Binary sessions
Manages provider clients and safely recreates failed binary analysis sessions.
src/application/binary/BinarySession.ts
Ghidra bridge
Exposes read-only Ghidra procedures, decompilation, references, and assembly through an authenticated bridge.
bridge/ghidra/ReaGhidraBridge.java
Generated