DECX - Decompiler + X

A JADX-based Decompiler + X - Designed for AI-assisted code analysis

Overview

DECX (Decompiler + X) is a smart code analysis platform built on the JADX decompiler, designed specifically for AI-assisted code analysis. The platform provides powerful Java code analysis capabilities to AI assistants through HTTP API and MCP (Model Context Protocol).

Installation

Prerequisites

• Java: JDK 11+
• Node.js: 18+ for the CLI
• JADX: v1.5.2+ with plugin support if you use the GUI plugin
• Python: 3.10+ or uv for the plugin MCP sidecar; without uv, ensure requests, fastmcp, and pydantic are available

CLI And AI Skills

For AI-assisted CLI work, install the CLI, install the DECX server JAR, then expose the bundled skills to your agent:

npm install -g @jygzyc/decx-cli
decx self install
git clone https://github.com/jygzyc/decx ~/.decx/source
mkdir -p ~/.agents
ln -s ~/.decx/source/skills ~/.agents/skills

Replace ~/.agents/skills with the skills directory expected by your agent:

The skills/ directory contains:

JADX Plugin

Install the plugin from the JADX GUI plugin manager, or install a plugin JAR manually:

jadx plugins --install-jar <path-to-jadx_decx_plugin.jar>

After installation, open an APK/JAR in JADX and enable DECX. The plugin exposes the DECX HTTP API and MCP tools for the currently opened JADX project.

Usage

CLI + Skills

For agent-driven analysis, use the CLI to create a session and let the installed skills drive the detailed workflow:

decx process open target.apk --name target
decx code classes --limit 50
decx code search-global "WebView" --limit 20
decx ard exported-components
decx ard app-deeplinks
decx process close target
decx process close --port 25419

Typical skill sequence:

• decxcli for exploration and evidence gathering
• decxcli-app-vulnhunt or decxcli-framework-vulnhunt for focused vulnerability hunting
• decxcli-poc for turning one confirmed finding into a buildable PoC

Useful command groups:

Notes:

• Session-backed code and ard commands support --page <n> and can target a session with -s, --session <name> or a port with -P, --port <port>.
• decx code class-source supports --limit <n> to return at most N source lines.
• decx process open <file> passes standard jadx-cli flags through and enables --show-bad-code by default.
• decx ard all-resources supports file-name filtering with --include and --no-regex.
• system-services and perm-info are adb-backed commands. They use --serial / --adb-path, not -P <port>.
• decx ard framework run collects from the connected device, processes, packs, and opens the final framework JAR by default; process <oem> is for local framework dumps.

Plugin + MCP

Use the plugin when you want the AI assistant to work against the project already opened in JADX GUI:

1. Open the target APK/JAR in JADX.
2. Enable the DECX plugin and confirm the server is available at http://127.0.0.1:25419.
3. Connect your MCP client to DECX and call health_check().
4. Use MCP tools for code search/source/xrefs, Android manifest/resources/components, framework service lookup, and JADX GUI selections.

All MCP tools support pagination with page where the returned content is large.

Plugin options you may need:

• decx.port: DECX HTTP server port, default 25419
• decx.mcp_path: custom MCP script path when you do not want the bundled script
• decx.cache: disk or memory, default disk

Error Codes

DECX returns the same structured error format from plugin and standalone server modes:

Error Response Format:

{
  "ok": false,
  "error": {
    "code": "CLASS_NOT_FOUND",
    "message": "Class not found: com.example.Foo"
  }
}

Development

Project Structure

Core request path:

CLI / MCP / HTTP
  -> DecxServer / RouteHandler
  -> DecxApi / DecxApiImpl
  -> service/* and utils/*

Build

cd decx
./gradlew dist

cd ../cli
npm install
npm run build
npm test

Contributing

1. Fork this repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a Pull Request

License

This project is licensed under GNU License - see the LICENSE file for details.

Credits

• skylot/jadx - The foundation of this project, a powerful JADX decompiler with plugin support
• zinja-coder/jadx-ai-mcp - Provided many ideas and inspiration, excellent practices for JADX MCP integration
• FastMCP: MCP protocol implementation
• Javalin: Lightweight web framework

⭐ If this project helps you, please give it a Star!