REA + Ghidra Toolkit

A small, dependency-light toolkit for authorized static analysis of native binaries with REA and Ghidra. It is intentionally built as a reusable knowledge base for agents, not as a dumping ground for a specific target.

Think in two layers

• This repository contains only generic methods, tools, and synthetic tests.

• Every target-specific artifact, decompilation, or private result stays in the private analysis project.

When this applies

The toolkit is aimed at native, unmanaged PE32 binaries on Windows:

• classic 32-bit Delphi applications have dedicated guidance (VMT / RTTI);

• other PE32 code (e.g. C/C++ components) works with the read-only inventory and the headless Ghidra export;

• analysis is static only; every claim is verified before it is treated as a finding.

Not a fit for managed runtimes (.NET / Java), PE32+ 64-bit images (the quick parser is deliberately PE32-only), executing code as a debugging workflow, or any target without solid legal authorization.

Related MCP server: erdgraph

Why this exists

You want to avoid rebuilding the same reverse-engineering workflow from scratch every time. This toolkit answers:

• How do I start a PE / Delphi analysis in a traceable way?

• How do I combine REA, Ghidra, and small local tools without treating inferences as facts?

• How do I export and document results so a later session can continue?

• Which mistakes should be avoided early instead of discovered late?

Repository layout

.
├── docs/
│   ├── workflow.md
│   ├── rea-ghidra.md
│   ├── delphi-pe32.md
│   ├── evidence.md
│   ├── troubleshooting.md
│   └── gitmcp.md
├── tools/
│   ├── pe32.py
│   ├── pe_inventory.py
│   └── delphi_vmt.py
├── ghidra/
│   └── scripts/ExportDecompilation.java
├── templates/
│   ├── analysis-log.md
│   └── handover.md
├── tests/
├── assets/
├── AGENTS.md
├── LICENSE
└── llms.txt

What you get

Quickstart

# From the repository root
python -m tools.pe_inventory "C:\analysis\program.exe"
python -m tools.pe_inventory "C:\analysis\program.exe" --imports --json

# Check a known VMT candidate
python -m tools.delphi_vmt "C:\analysis\program.exe" --vmt 0x00412340

# Heuristic search; always validate candidates in a disassembler
python -m tools.delphi_vmt "C:\analysis\program.exe" --scan

Run the synthetic test suite:

python -m unittest discover -v

GitMCP

This repository is designed to be used with https://gitmcp.io/ethanhunt1981/rea-ghidra-toolkit.

It is also listed in the official MCP Registry as io.github.ethanhunt1981/rea-ghidra-toolkit (streamable HTTP via GitMCP; metadata in server.json).

OpenCode / typical agent setup:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "servers": {
      "rea-ghidra-toolkit": {
        "type": "remote",
        "url": "https://gitmcp.io/ethanhunt1981/rea-ghidra-toolkit"
      }
    }
  }
}

OpenCode CLI command:

opencode mcp add rea-ghidra-toolkit --global --url https://gitmcp.io/ethanhunt1981/rea-ghidra-toolkit
opencode mcp list

Core rule

Observation, inference, and unknown are three different things. A plausible name, string, or VMT candidate is not yet a behavioral proof. The templates are here to keep that distinction visible across sessions.

License

This repository is licensed under the MIT License. The license covers the included methodology and tooling only. It does not cover original binaries, target-specific specifications, third-party protected material, or private analysis results.