Ghidra Headless
Provides a headless Ghidra server to equip AI agents with deep reverse-engineering capabilities.
A headless Ghidra server that speaks Model Context Protocol (MCP), specifically designed to empower AI agents with comprehensive reverse-engineering capabilities. It enables agents to automate complex workflows like program lifecycle management, disassembly, decompilation, patching, and scripting, all without a graphical user interface. Optimized for sandboxed VM/container environments, it allows agents to inspect, refine, and extend analyses incrementally, improving analysis databases and iterating safely through transactions and custom scripts.
主な機能
01Provides scripting access via `ghidra.eval`, `ghidra.call`, and `ghidra.script` for custom workflows.
02Features a real `pyghidra` backend for live headless Ghidra analysis and a fake backend for CI/development.
0334 GitHub stars
04Exposes 212 tools across 34 feature groups for comprehensive reverse engineering.
05Defaults to read-only mode with safe mutation workflows using transactions, undo/redo, and explicit saves.
06Supports both Stdio and TCP transports for flexible agent communication.
ユースケース
01Enabling AI agents to iteratively refine and extend Ghidra analysis databases over time.
02Co-locating the Ghidra server with AI agent runtimes within the same Docker container for efficient deployment.
03Automating extensive reverse engineering tasks for AI agents in sandboxed environments.