zinja-coder/jadx-ai-mcp

📦 Project: JADX‑AI‑MCP (part of the Zin MCP Suite)

What it is

• A plugin for the popular JADX decompiler that talks to a Model Context Protocol (MCP) server. The server forwards requests to an LLM (e.g., Anthropic’s Claude) so the model can inspect the decompiled Android app in real‑time.
• Think of it as a live AI‑powered code reviewer for Android APKs: you select a class, method, or resource in JADX, the LLM receives the context, and it replies with explanations, vulnerability checks, refactorings, or even automatic renaming.

Key components

Component	Role
JADX‑AI‑MCP (Java plugin)	Runs inside the JADX GUI, gathers source/SMALI, sends MCP requests, displays AI responses as comments or UI dialogs.
JADX‑MCP‑Server (Python)	Stand‑alone MCP server that implements the protocol, talks to the LLM client (Claude Desktop, LM‑Studio, Cherry Studio, etc.), and forwards data to/from the plugin.
MCP (Model Context Protocol)	A lightweight JSON‑over‑stdio/HTTP protocol that lets any LLM client treat the decompiled app as a “context” it can query.

What you can ask the AI (sample prompts from the README)

• Code understanding: “Explain what this class does.”
• Vulnerability detection: “Are there insecure API calls here?”
• Reverse‑engineering helpers: “Deobfuscate and rename this class.”
• Static analysis: “List all network calls in this file.”
• Code modification: “Add Javadoc comments to every method.”
• Debugger assistance: “Fetch stack frames and summarise the execution flow.”

How to get started

1. Download the latest release – a jadx‑ai‑mcp‑<ver>.jar and a jadx‑mcp‑server‑<ver>.zip.
2. Install the plugin
   • One‑liner: jadx plugins --install "github:zinja-coder:jadx-ai-mcp"
   • Or drop the JAR into the JADX‑GUI plugin folder (screenshots in the README).
3. Set up the server
   • Install uv (a modern Python package manager) and run uv run jadx_mcp_server.py (optionally with --http for network transport).
4. Configure your LLM client (Claude Desktop, LM‑Studio, Cherry Studio, etc.) to point to the MCP server using the JSON snippet shown in the README.
5. Open an APK in JADX, select a piece of code, and type a prompt – the AI’s answer appears directly in the GUI.

Technical stack

• Java 11+ for the JADX plugin.
• Python 3.10+ for the MCP server (uses httpx and fastmcp).
• MCP (Model Context Protocol) – a language‑agnostic bridge between tools and LLMs.
• Works with any LLM that implements MCP (Claude Desktop is the primary example).

Maturity & status

• Early‑stage, actively developed; the README warns about bugs and crashes.
• Provides a full set of MCP commands (fetch class source, rename symbols, list resources, X‑ref queries, debugger introspection, etc.).
• Documentation hosted on Read‑the‑Docs and several demo videos are linked.

Who it’s for

• Security researchers and reverse‑engineers who want AI assistance while analysing Android binaries.
• Developers interested in rapid static analysis, deobfuscation, or automated code review of APKs.
• Anyone comfortable running a local Python server and a Java‑based decompiler.

---

In short, JADX‑AI‑MCP brings LLM‑driven, context‑aware analysis to the Android reverse‑engineering workflow, turning the decompiler into an interactive AI assistant.