Jadx AI MCP, by Zinja Coder, is an MCP server that extends the JADX Android decompiler with large language model capabilities. This free, open-source developer tool acts as a plugin for JADX, exposing its functions through the Model Context Protocol to AI clients. It requires Java 11+ for its plugin and Python 3.10+ for its companion server, which uses uv for dependency management. Jadx AI MCP is designed to assist with Android application analysis by enabling AI-driven queries against decompiled code.

What does Jadx AI MCP connect to?

Jadx AI MCP functions as a plugin for the JADX Android decompiler, establishing a direct connection to its core functionalities. It then exposes these capabilities via the Model Context Protocol (MCP) to various large language model clients. This setup allows AI clients, such as Claude (tested with Claude Desktop) and Cursor, to interact directly with JADX's decompiled output. The system supports both HTTP and stdio transport protocols for communication between the MCP server and connected AI clients.

Core features and tools

Jadx AI MCP provides several callable tools for interacting with decompiled Android applications. Key functionalities include fetch_current_class for selected class source, get_selected_text for current text, and get_all_classes to list project classes. It offers get_class_source(class_name), get_method_by_name(class_name, method_name), and get_smali_of_class(class_name). Search capabilities include search_method_by_name(method_name) and search_classes_by_keyword(keyword). Cross-referencing tools are xrefs_to_method() and xrefs_to_field(). It also exposes get_android_manifest.

Who is it for?

Jadx AI MCP targets security researchers conducting mobile app assessments and Android developers analyzing third-party libraries. DevSecOps teams can use it for mobile security scanning, and it serves anyone needing to understand Android applications. It aids in automating reverse engineering tasks and integrating AI code analysis into security audits. This is not the right fit if your workflow does not involve Android application decompilation or AI-driven code analysis.

How to get started

No credentials are required when the MCP server is configured to bind to all interfaces using --host 0.0.0.0. The system supports HTTP and stdio transport protocols for communication.

Use cases

• Automating reverse engineering tasks by querying decompiled code via AI.
• Integrating AI code analysis into Android app security audits.
• Building custom analysis pipelines that combine JADX output with AI reasoning.

Jadx AI MCP by Zinja Coder extends JADX with AI capabilities for Android application analysis. It provides a set of tools for querying decompiled code, supporting various security and development workflows. The system integrates with LLMs like Claude and Cursor, enabling AI-powered insights into Android binaries. Its unauthenticated HTTP access simplifies initial setup for local use. Try using AI to "analyze this APK for hardcoded credentials" to begin exploring its capabilities.