It does a single thing — hand a prompt to the Gemini command-line tool and bring the reply back — which is enough to get a second read on a problem without switching windows. Because it drives that tool rather than the API directly, whatever the tool can already do on your machine is what you get, including the variant that skips its confirmation step.
A small stdio server that drives Google's Gemini command-line tool. It passes your prompt to that tool and returns the reply, so a second model can be consulted mid-task without leaving the conversation.
- A prompt sent to Gemini and its answer returned in the conversation
- A second variant that auto-confirms the operations the command-line tool would otherwise stop and ask about
- A stdio transport, so it connects the way any local server does
Node on your machine, Google's Gemini command-line tool installed, and a Gemini API key supplied through the environment. You clone the repository and build it yourself, then point your client at the built entry file.