tasopen/mcp-alphabanana — reverse-engineered prompt

Reverse engineered prompt

Build me a local MCP server that lets Claude Desktop, Cursor, or another MCP client generate image assets with Google Gemini. I want one main tool where I can type a prompt and get an image back, with options for a fast model or a higher quality model, optional local reference images to guide the style, and the ability to save the result to a folder or return it as base64 or both.

Please support common asset workflows like exact width and height resizing, keeping native aspect ratio when I want that, and transparent background output for PNG or WebP so it works well for web or game assets. It should also let me choose jpg, png, or webp, and handle local file paths cleanly. Make it easy to run with an API key from an environment variable, and make sure it works nicely as an MCP server people can hook into their client config.

If you need details, look up the current Gemini image and MCP docs online and wire it up in a clean, usable way.

Want more depth? Deep Reverse