JADX-AI-MCP: A Plugin That Lets Claude Read Your Decompiled Android Code
Plugin for JADX to integrate MCP server
At a glance
- What is it?
- A JADX plugin that exposes decompiled APK data to LLMs through the Model Context Protocol. It promises live reverse engineering assistance, but the documentation is still thin in places.
- Who is it for?
- Adopt JADX-AI-MCP if you are a mobile security analyst or reverse engineer who already works in JADX and wants an LLM like Claude to query decompiled code without copying snippets manually. Skip it if you need a fully documented, stable tool for production pentesting or if you cannot accept the risk of an LLM misreading obfuscated logic.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 17 days ago.
- What is it written in?
- Mainly Java, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What the Plugin Actually Does
JADX-AI-MCP is a plugin for the JADX decompiler that adds a Model Context Protocol (MCP) server. The MCP server is the bridge that lets an LLM such as Claude talk to the decompiled code inside JADX. The README describes the goal as 'live reverse engineering support with LLMs.' Instead of copying a decompiled method into a chat window, you ask the model questions about the code that is already open in JADX. The intended users are security researchers, pentesters, and anyone doing static analysis of Android APKs. The project tags include 'pentesting', 'sast', and 'vapt', so the focus is vulnerability discovery and code review, not casual exploration. The plugin is part of a larger 'Zin MCP Suite', which suggests the author plans a family of MCP-enabled tools. The repository is not archived, and the latest release is V6.4.1 from August 2026, so development is ongoing.
How the MCP Integration Works
The architecture is a JADX plugin that also runs an MCP server. MCP, or Model Context Protocol, is a standard from Anthropic for giving LLMs access to external tools and data. The plugin likely registers tools that the LLM can call, such as 'get method source' or 'find references', but the README does not list specific tool names. What is clear is that the plugin runs inside JADX, so the decompiled code is the context. The LLM does not see the raw APK; it sees the decompiled Java source that JADX produces. That distinction matters because JADX's output is already a reconstruction, not the original code. The README mentions 'real-time code review', which implies that as you navigate JADX, the plugin can feed current selections or the whole class to the model. However, the documentation is vague on the exact data flow. There is a ReadTheDocs site linked, which likely has the details, but the repository README itself is more marketing than specification.
Getting It Running: Installation and Setup
The README does not give explicit installation commands, which is a gap. It states that Java 11+ and Python 3.10+ are required. The Python requirement is interesting because the plugin is Java-based; likely the MCP server or a helper component uses Python. The project has releases, so you can download a release artifact, probably a JAR or a zip, from the GitHub releases page. The default branch is named 'jadx-ai', which suggests you may need to build from source if you want the latest. To build a JADX plugin, you typically clone the repository and use Gradle, but the README does not show the command. The presence of a ReadTheDocs site means the actual setup is probably documented there. For a tool that promises 'fully automated' integration, the lack of a quick start in the main README is a real friction point. You will need to consult the external documentation or figure out the build process yourself.
Limitations and Failure Modes
The README contains a comment that says the project is 'still in early stage of development, so expects bugs, crashes and logical errors.' That is a direct warning. The plugin is a bridge between a decompiler and an LLM, and both sides can fail. JADX itself can produce incorrect decompilation for obfuscated code, and the LLM will happily reason about that incorrect code. The MCP server may crash if the APK is very large or if the model makes too many requests. The remote host support added in v6.3.0 suggests you can run the server on a different machine, but that also introduces network latency and authentication concerns. The plugin is wrong tool for automated, large-scale analysis where you need deterministic results. An LLM is probabilistic, so you cannot trust it to find every vulnerability. It is best as an assistant to a human analyst, not as a replacement for static analysis tools like semgrep or MobSF.
Real Alternatives: JADX Alone or Other MCP Servers
The obvious alternative is to use JADX without the plugin and manually copy code into an LLM chat. That approach gives you full control and no extra moving parts, but it is slow and loses context. Another alternative is a standalone MCP server for code analysis that does not tie to JADX. For example, you could use a generic MCP server that reads files from disk and point it at JADX's exported source. That approach is more flexible because it works with any decompiler, but it lacks the live integration. The key difference is that JADX-AI-MCP operates on the in-memory decompilation inside JADX, which means it can follow your navigation and selections. A file-based server has to re-index the exported source. If you are already using JADX daily, the plugin is more convenient. If you need to analyze code from multiple tools, a generic server is better.
Maintenance, Licensing, and Upgrade Cost
The project is under Apache-2.0, which is permissive and allows commercial use, but the README does not state that you must attribute the original authors. That is not legal advice; check the license file. The release history shows a steady cadence: v6.3.0 in March 2026, v6.4.0 in May, and v6.4.1 in August. That suggests active maintenance, but it also means you will need to keep up with releases to get bug fixes. The v6.4.0 release notes mention 'Search Infrastructure Overhauled', which is a sign that core components change between versions. Upgrading may require re-testing your workflow. The plugin depends on JADX's internal APIs, and JADX itself changes, so an upgrade of JADX could break the plugin. You should pin both JADX and the plugin versions. The Python dependency adds another upgrade path. There is no mention of a changelog or migration guide in the README, so expect to read release notes on your own.
Editorial conclusion
Adopt JADX-AI-MCP if you are a mobile security analyst or reverse engineer who already works in JADX and wants an LLM like Claude to query decompiled code without copying snippets manually. Skip it if you need a fully documented, stable tool for production pentesting or if you cannot accept the risk of an LLM misreading obfuscated logic. Before adopting, verify that your JADX version matches the plugin's expected build, check the ReadTheDocs for the current setup steps, and test on a sample APK to see how the MCP server handles large codebases. The project is active and Apache-2.0 licensed, but its early-stage warnings mean you should not rely on it as your only analysis tool.
Community notes