jadx-ai

JADX-MCP-SERVER is a Python-based MCP server that exposes a live decompiler context from JADX to language models via the MCP protocol. It enables real-time, context-aware analysis and code-review prompts by exposing endpoints and tools that let an LLM query the currently selected class, entire project structure, or specific class/method sources. Available MCP tools include fetch_current_class, get_selected_text, get_all_classes, get_class_source, get_method_by_name, search_method_by_name, get_methods_of_class, get_fields_of_class, get_method_code (deprecated alias), and get_smali_of_class. This setup allows an LLM to inspect the decompiled Android project, fetch sources on demand, and provide inline recommendations or refactors during reverse engineering or static analysis sessions. To use it, run the MCP server with uv, ensure the JADX GUI is open and a class is selected, and then integrate the server into your MCP-enabled client (e.g., Claude Desktop) to issue the prompts described in the sample prompts section of the README.

Prerequisites:

• Python environment (3.8+ recommended)
• uv (Astral's uv) as the dependency manager
• Access to jadx-mcp_server folder with jadx_mcp_server/jadx_mcp.py and requirements.txt

Installation steps:

1. Install uv if you don’t have it: curl -LsSf https://astral.sh/uv/install.sh | sh

2. Create and activate a virtual environment (optional but recommended): uv venv source .venv/bin/activate # On Windows use .venv\Scripts\activate

3. Install MCP server dependencies: uv pip install httpx fastmcp

4. Ensure you have the jadx-mcp_server folder available from the releases (jadx_mcp_server/jadx_mcp.py and requirements.txt).

5. Run the MCP server: uv --directory </PATH/TO/>jadx_mcp_server/ run jadx_mcp.py

6. (Optional) Configure Claude Desktop or your MCP client to point at the server using the provided example configuration in the README, replacing the directory path with your actual path.

Notes:

• The server is designed to run alongside JADX-GUI and expects a live GUI session for context.
• If you modify paths, ensure they’re accessible by the user running uv.

Tips and common issues:

• Ensure the jadx-gui is running and a class is actively selected when testing MCP tool prompts.
• If you encounter network or HTTP errors, verify that the server is reachable from the MCP client and that the port/endpoint configurations (if customized) are correct.
• Use the sample prompts in the README to validate that the MCP tools return expected sources and metadata.
• The environment variable UV_DIRECTORY (or equivalent) should point to the directory containing jadx_mcp_server so the server can load modules correctly.
• This is a developer beta; expect bugs and edge cases during reverse-engineering workflows.