An MCP server for Claude Code, Codex, Cursor and other agents

See a feature you like. Trace how it works, down to the binary.

REA lets your coding agent inspect apps, binaries, captures and live behaviour on your machine, then hands back findings with the evidence attached: addresses, files and references.

See how it works

Needs Node.js 22, 24 or 26. Runs locally; nothing is uploaded by REA.

sound.dll pan_from_x

  1. 00401a2055 push ebp
  2. 00401a218b ec mov ebp, esp
  3. 00401a238b 45 08 mov eax, [ebp+8]
  4. 00401a262d 40 01 00 00 sub eax, 0x140
  5. 00401a2b69 c0 10 27 00 00 imul eax, eax, 0x2710
  6. 00401a3199 cdq
  7. 00401a32b9 40 01 00 00 mov ecx, 0x140
  8. 00401a37f7 f9 idiv ecx
Evidence at 00401a26: 0x140 is the screen centre, 320.Illustrative listing
target types, from ELF to APK
42target types, from ELF to APK
native engines: Hopper, Ghidra, IDA, Kuna
4native engines: Hopper, Ghidra, IDA, Kuna
MCP server for every agent
1MCP server for every agent
uploads: analysis stays local
0uploads: analysis stays local

Your agent asks. REA collects the evidence.

The same workflows run from the command line, so anything your agent finds you can reproduce yourself.

  1. You ask

    Point your agent at a local app, binary, capture or site and ask how a feature works.

  2. REA inspects

    The agent calls REA over MCP. REA drives the right analysis tool and traces the relevant code.

  3. Evidence comes back

    Findings arrive with addresses, files and references attached, plus what is still unknown.

  4. Your agent builds

    It explains the behaviour, or writes and tests an implementation for your project.

What you can point it at

Most formats are read statically with nothing extra installed. Where REA needs another tool, it says which.

42 of 42 target types

  • Native binaries

    Pseudocode, assembly, strings, symbols, calls and references

    Hopper, Ghidra, IDA or Kuna

  • Desktop installations

    Runtime candidates, supporting files, component markers and next probes

    Built in

  • Offline ELF layout

    Sections, segments, symbols, relocations and mitigation candidates

    pwntools on Linux x64

  • Go binaries

    Compiler version, modules, replacements and build settings

    Built in

  • Rust binaries

    cargo-auditable crate names, versions, sources and dependency edges

    Built in

  • GPU code objects

    CUDA/HIP targets, kernel parameters, resources and PTX text

    Built in

  • PE certificates

    Authenticode entries and PKCS#7 signer identity fields

    Built in

  • PE imports

    Import and delay-load DLLs, ordinals, IAT RVAs and import hash

    Built in

  • ELF notes

    Note owners, types, GNU build-id and ABI-tag fields

    Built in

  • ELF dependencies

    DT_NEEDED, SONAME, RPATH/RUNPATH, versions and undefined imports

    Built in

  • DWARF compile units

    Unit name, directory, producer, language and dwo_name

    Built in

  • Binary hardening

    PIE, NX, RELRO, CFG and related flags for ELF, PE and Mach-O

    Built in

  • Windows shortcuts

    Shell Link target, arguments, working directory and tracker data

    Built in

  • Linux kernel modules

    .modinfo keys, __versions CRCs and signature trailer

    Built in

  • Apple IPS crashes

    Threads, images, registers and documented crash classes

    Built in

  • PyInstaller executables

    Archive members, Python version, entry scripts and bytecode

    Built in

  • Deno compile binaries

    Virtual files, entry point, options and npm snapshot length

    Built in

  • Firmware

    Regions, extraction results and native-analysis handoffs

    Binwalk or Unblob on Linux

  • .NET assemblies

    Metadata, CIL instructions, native dependencies and build comparisons

    Built in

  • WASM artifacts

    Validation, sections, import/export forms and decoded WAT

    WABT 1.0.42

  • EVM bytecode

    Dispatch selectors, offsets, inferred arguments and mutability

    Linux x64

  • Erlang BEAM

    Chunks, atoms, exports, imports, lambdas and code header

    Built in

  • IL2CPP metadata

    Unity assembly images, type and method names and index ranges

    Built in

  • Godot PCK

    Archive inventory, extraction and header facts

    Built in

  • Model artifacts

    GGUF and safetensors metadata, tensors and derived totals

    Built in

  • NuGet packages

    Identity, dependency groups, licence, repository and contents

    Built in

  • Packages and resources

    Inventories, digests, plists, Apple bundle anatomy and resources

    Built in

  • Android APKs

    Manifest declarations, classes, decompiled methods and references

    JADX and a full JDK

  • APK signatures

    v1–v3.1 signers, certificate fingerprints and key rotation

    Built in

  • Android devices

    Packages, processes, logs, dumps, screen captures and transfers

    adb

  • Android resources

    Decoded manifest, string tables, locales and version facts

    Apktool

  • Flutter apps

    Dart snapshot hash, per-ABI library digests and engine provenance

    Built in

  • JEB projects

    Project units, artifact digests and type/method pseudo-code

    A JEB client serving MCP

  • JavaScript and Electron

    Modules, imports, source maps, routes, IPC and native add-ons

    Node.js and npm

  • Websites

    Page structure, scripts, network observations and screenshots

    A Chrome-family browser

  • Saved network captures

    Requests, responses, exposed payloads and source locations

    HAR or Fiddler SAZ

  • Process behaviour

    Terminal output, interactions, exit and filesystem observations

    Linux or macOS PTY

  • SQLite databases

    Tables, columns, indexes, views, triggers and selected rows

    Node.js 24.15+

  • Recorded Linux crashes

    Notes, every thread's registers and signals, mapping candidates

    pwntools on Linux x64

  • Recorded packets

    Selected fields, repeated values, filters and decode-as rules

    TShark 4.x on Linux

  • Packet summaries

    Endpoints, DNS, TLS SNI, protocol mix and HTTP requests

    tshark and capinfos

  • Recorded logic signals

    Decoder annotations, sample ranges and diagnostics

    sigrok-cli 0.7.2 on Linux

Worked investigations

Each case starts with a plain question and ends with a reconstruction you can check.

Set it up in one command

Setup shows every change before it writes anything, and keeps backups of your agent's configuration.

Choose your agents, review the proposed changes and approve them. Setup registers REA's MCP server and its workflow instructions, backing up your existing configuration. Restart your agent afterwards.

npx rea-agents setup

Then ask your agent something like: “Understand how search works in the Notes app, show me the evidence, and build a similar feature for my project.”

Questions

Which agents can use REA?

Any agent that supports local MCP servers. Setup configures Claude Code, Codex, Cursor, Gemini CLI, Grok Build and others; any other client can register the MCP server by hand.

Do I need Hopper, Ghidra or IDA?

Only for deep native analysis. JavaScript, Electron, .NET and most file formats are inspected without one. Setup can install Hopper after you approve it; Ghidra, IDA and Kuna use your existing installs.

Does REA upload my app?

No. REA analyses targets on your machine. Your agent receives the tool results, and its model provider's own data policy applies to those.

What code does REA give back?

Native analysis returns pseudocode and assembly; JavaScript analysis recovers modules and how they connect. Your agent uses those findings to write and test its own implementation.

Which Node.js version do I need?

Node.js 22.x (22.23.4 or later), 24.x (24.22.1 or later) or 26.12.1 and up, with npm.

Something broke. What now?

Update first with `rea update` or `npx rea-agents@latest setup`, restart your agent and retry. If it still fails, open an issue on the REA repository with your version, target type, steps and error output.