mrexodia/ida-pro-mcp

IDA Pro is a professional tool that security researchers and reverse engineers use to read compiled programs, the kind of software where you only have the finished binary file and no human-readable source code. This project adds an AI assistant layer on top of IDA Pro by connecting it to language models through a protocol called MCP (Model Context Protocol). The result is that an AI can look at the decompiled output, suggest better names for variables and functions, add explanatory comments, and help reason through what a piece of code is actually doing. To use it, you need a paid copy of IDA Pro version 8.3 or higher and Python 3.11 or higher. You also need one of the supported AI clients, which include Claude, Cursor, Copilot, VS Code, and more than a dozen others. Installation runs through a couple of command-line steps: you install the package with pip and then run a single command that sets up the plugin inside IDA Pro and configures the MCP connection for your chosen client. The README is honest about where AI falls short in this context. Language models can make up facts, and they handle number conversions poorly, so the project provides a dedicated tool called int_convert to handle that and recommends pairing it with a math helper for other calculations. Obfuscated code is another weak spot: things like encrypted strings, scrambled control flow, or anti-decompilation tricks should be cleaned up before asking an AI to analyze anything, because the model will not perform well on them. Beyond the standard plugin mode, there is a headless option called idalib-mcp. This lets you run the analysis server without the IDA graphical interface at all, which is useful for automated pipelines. The headless server can manage multiple binary databases at once, each in its own background process, and can open and close files dynamically over its lifetime. The project includes example prompts for guiding the AI through a reverse engineering session, such as analyzing a crackme puzzle or producing a structured report of findings. These are shared as starting points rather than definitive recipes, and the README invites users to share better approaches.