`jpeg_quality`, `webp_quality`, `png_compression_level`, `optipng_optimization_level` and `target_image_max_size` all shape what is written to disk, and `skip_compression_and_resizing` turns that stage off in one flag — so the image arrives already sized for wherever it is going, rather than needing a second pass. The vendor's description of the tool is a single sentence, which makes the argument list the real documentation. With a single tool there is no call ordering to learn.
A single-tool image generation server: `generate_image` takes a `prompt` and saves the result into `output_directory` under `file_name`.
- A single call that generates the image and writes the file: `prompt` carries the request, `output_directory` and `file_name` decide where it goes.
- Output handling in the same arguments: `force_conversion_type` for the format, `jpeg_quality`, `webp_quality`, `png_compression_level` and `optipng_optimization_level` for the encoders, `target_image_max_size` for the ceiling, and `skip_compression_and_resizing` to bypass that stage.
- `input_image_paths` for sending existing images along with the request, and `use_enhanced_prompt` for switching prompt enhancement on.
A Gemini API key, supplied as GEMINI_API_KEY, and a writable `output_directory` for the file the call produces.
One command plus a key — npx -y @creating-cat/gemini-image-mcp-server, then supply credentials
