elliottophellia/Ryuumonbuchi ? reverse-engineered prompt

Build me a headless Ghidra MCP server in Python that I can point at a binary and use from an LLM agent to analyze it without opening the Ghidra GUI.

I want to be able to start the server locally, open a program, run decompilation, disassembly, symbol and type edits, memory reads and writes, patching, and save or export the project when allowed. It should keep one warm Ghidra session around so repeated tool calls are fast, and it should support batching multiple analysis actions in one request. Please make it work over MCP, with stdio as the default and HTTP as an option. Use sensible defaults, but let me configure the Ghidra install path, timeouts, memory limits, and export or import permissions from environment variables or command line flags.

If anything is unclear, look up the current Ghidra and MCP docs online and follow the repo’s README behavior closely.