Native MCP Server Extension for Ghidra

[betab0t]

Hi Ghidra team,

I am exploring the idea of developing a native Model Context Protocol (MCP) server for Ghidra to enable programmatic access for AI agents/assistants and automated analysis workflows. There are a few community plugins that implement this, but none are integrated into upstream. I would like to understand the feasibility before investing in a full implementation.

Specifically:

1. Would the team be open to reviewing and potentially upstreaming a self-contained MCP server module that lives in Extensions/ and is completely optional when disabled?
2. Are there any technical, security, or architectural constraints I should know about before starting?
3. Is there anything I can do to make this easier for the team, such as scope boundaries, testing expectations, or API areas to avoid?
4. Is there any internal work or direction around remote-control or automation APIs that I should align with?

My goal is to keep the module clean, isolated, and non-intrusive so it can be used as a standalone plugin and also qualify as a potential upstream contribution if you'll find it fit.

Thanks for your time. Any guidance would be appreciated.

[betab0t]

For reference, I am aware of the existing community MCP projects:

• GhidraMCP - Has not been updated for a while, and is missing interfaces for defining functions and creating struct definitions, etc'.
• pyghidra-mcp (Python) - Not a native implementation (python). Works only in headless mode so it cannot support interactive users.

Both projects are great, but with the way AI agents are evolving today, we need something more solid and fully integrated.