radareorg/radare2-mcp

What it solves

It allows AI agents to interact with radare2, a powerful open-source reverse engineering and binary analysis framework. By providing a standardized interface, it enables LLMs to perform complex binary analysis tasks directly through agents like Claude, VS Code Copilot, or Zed.

How it works

The project implements a Model Context Protocol (MCP) server written in C using native radare2 APIs. It can operate in three modes: as a standalone CLI tool using stdin/stdout communication, as an HTTP server for remote or multiplexed sessions, or as a core radare2 plugin that integrates directly into an existing analysis session.

Who it’s for

Security researchers, reverse engineers, and developers who want to leverage AI agents to automate or assist in the analysis of compiled binaries.

Highlights

• Flexible Deployment: Works as a CLI, an r2 plugin, or an HTTP server.
• C-Native Performance: Written in C using native r2 APIs for seamless integration.
• Secure Access: Supports readonly mode, sandbox locks, and restricted toolsets to prevent unauthorized actions.
• Broad Compatibility: Integrates with major AI clients including Claude Desktop, VS Code (GitHub Copilot), and Zed.
• Extensible: Provides optional raw access to execute r2 commands or r2js scripts.