ReVa
A Ghidra MCP server: the agent decompiles, renames variables, builds cross references and works through a binary via a set of small tools
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- Exposes the PyGhidra environment, meaning arbitrary code execution
- A reverse engineering tool; use it on your own or authorized binaries
- A public interface without an API key grants access to analysis and scripting
Install
Manual install
claude mcp add --scope user --transport http ReVa -- http://localhost:8080/mcp/messageAssistant mode: Ghidra must be open with a project.
This is third-party code. Review the repository files before installing.
What it does
ReVa is a Ghidra extension that runs an MCP server and exposes Ghidra to an AI agent. Its approach is built on many small tools: decompilation with context about names and references, variable renaming, type work, string search and diffing two programs. This set reduces context use and helps the model explore a binary step by step, the way a human does. It works in two modes: an assistant inside the Ghidra UI over streamable HTTP on port 8080 and headless for automation and pipelines. The repository ships a Claude Code plugin marketplace with skills for triage, deep analysis, cryptography and CTF.
Who it is for. For reverse engineers and analysts who work in Ghidra and want to connect an agent to the analysis.
Good fit when
- You run analysis in Ghidra and want an assistant in the same project
- You need headless mode for automation and pipelines
- You work through a large binary or firmware and context economy matters
Not a fit when
- You analyze someone else's proprietary code without rights
- Your Ghidra is below version 12.0
- The server is exposed on a public interface without an API key and with scripting enabled
Example request
Start from main, examine the program in detail and rename variables as you goLimitations
It needs Ghidra 12.0 or newer, and headless mode uses the GHIDRA_INSTALL_DIR variable. ReVa exposes the PyGhidra environment, so when listening on a public interface the authors advise enabling an API key or disabling the scripting tools. By default the server listens on localhost. Headless projects are ephemeral and cleaned up after the session.
How to disable. Disable the ReVa plugins in Ghidra settings and remove the MCP entry from the client. The headless package is removed with uv tool uninstall reverse-engineering-assistant.
MCP
- Transport
- http, stdio
- Authentication
- API key
| Environment variables | |
|---|---|
| GHIDRA_INSTALL_DIR | Path to the Ghidra install, required for headless mode. |
Security check
- Exposes the PyGhidra environment, meaning arbitrary code execution
- A reverse engineering tool; use it on your own or authorized binaries
- A public interface without an API key grants access to analysis and scripting
README in short
The README explains ReVa's small-tools approach that reduces context use and gives example questions about a binary. It describes installing the Ghidra extension and enabling two plugins, and two modes: assistant over streamable HTTP on port 8080 and headless for automation. It gives config for Claude Code and VS Code, PyGhidra integration and installing the Claude Code plugin marketplace with skills. It notes the Ghidra 12.0 requirement and a warning about the API key on a public interface.
FAQ
How does assistant mode differ from headless?
In assistant mode ReVa runs inside an open Ghidra project over HTTP. In headless it starts Ghidra itself and creates an ephemeral project, which suits automation.
Are there Claude Code skills?
Yes. The repository has a plugin marketplace with skills for triage, deep analysis, cryptography and CTF.
Related
A code security audit skill by Cloudflare: the agent runs recon, coverage-led hunting and independent verification of findings, then produces a structured repor
NVIDIA's open stack for running OpenClaw, Hermes and LangChain Deep Agents in OpenShell sandboxes with network policy and managed inference
Security scanner for agent skills and MCP servers: finds prompt injection, data exfiltration and supply chain risks before install
Static code analysis with rules that look like source code, plus a built-in MCP server for AI agents