zinja-coder/jadx-ai-mcp

What it solves

JADX-AI-MCP allows users to perform live, AI-powered reverse engineering of Android APKs. It bridges the gap between a decompiler (JADX) and Large Language Models (LLMs) like Claude, enabling users to analyze decompiled code, find vulnerabilities, and deobfuscate code through a conversational interface without manually copying and pasting code into an AI chat.

How it works

The project consists of two main components that communicate via the Model Context Protocol (MCP):

1. JADX-AI-MCP Plugin: A Java-based plugin installed directly into the JADX GUI that gathers data from the decompiled app and performs actions (like renaming variables).
2. JADX MCP Server: A Python-based standalone server that acts as the intermediary. It receives tool requests from an MCP-compatible LLM client (like Claude Desktop) and translates them into HTTP requests for the JADX plugin.

Who it’s for

This tool is designed for security researchers, reverse engineers, and Android developers who need to analyze APKs for vulnerabilities, understand complex decompiled logic, or perform static analysis on Android applications.

Highlights

• Deep JADX Integration: Provides tools to fetch current classes, selected text, all classes, and specific methods or fields.
• Android Specifics: Can retrieve the AndroidManifest.xml, fetch main activity classes, and analyze APK resources and strings.
• **AI-Driven Refactoring: Supports renaming classes, methods, fields, packages, and variables directly via AI prompts.
• Debugger Support: Can fetch stack frames, threads, and variables from the JADX debugger for real-time execution flow analysis.
• Cross-Reference Analysis: Includes tools to find all references (xrefs) to specific classes, methods, or fields.