Connect ModelBrain to Gemini CLI

This client needs the full path to ModelBrain's mcp_server program. The commands and snippets on this page show the macOS path; on Windows or Linux, put the path for your system in the same place. Write the full path rather than a variable such as %LOCALAPPDATA% or ~, because a client may not expand it.

1. Install ModelBrain if you haven't yet (macOS, Windows or Linux).
2. Open ~/.gemini/settings.json. Run Gemini CLI once first if the file does not exist yet.
3. Add ModelBrain inside the mcpServers object. The file usually already holds other settings, such as a security block that Gemini CLI writes on first run. Keep them, and put a comma after the last one before your new block. Keep any servers already listed, and do not add a trust line.

   {
     "security": {
       "auth": {
         "selectedType": "oauth-personal"
       }
     },
     "mcpServers": {
       "modelbrain": {
         "command": "/Applications/ModelBrain.app/Contents/MacOS/mcp_server"
       }
     }
   }

   The security block above is only an example of an existing setting. Leave whatever your own file has.
4. Check the file is valid JSON. In a terminal run python3 -m json.tool ~/.gemini/settings.json. It prints the file back if it is valid and an error with a line number if it is not.
5. Check it is connected. In a terminal run gemini mcp list. ModelBrain should show as Connected. Start a new Gemini CLI session afterwards so it picks the change up.