The multimodal input is the reason to install this rather than call the API yourself: video, audio, PDFs and images go in by path with the MIME type worked out from the extension, so "summarise this recording" or "read this scanned document" is a single call with a file argument. The three capability switches are per call rather than per install, so the same tool serves a grounded question and an image generation without anything being reconfigured between them. The file-count and total-size ceilings are set by environment variables, which is where a large batch will stop.
A single-tool Gemini bridge: one generate_content call carrying a prompt, optional file inputs, and per-call switches for code execution, image generation and Google search.
- A user prompt with an optional system prompt, temperature, model and thinking budget
- File inputs with the MIME type detected from the extension: images (JPG, PNG, GIF, WebP, SVG, BMP, TIFF), video (MP4, AVI, MOV, WebM, FLV, MPG, WMV), audio (MP3, WAV, AIFF, AAC, OGG, FLAC), PDFs, and text formats (TXT, MD, JSON, XML, CSV, HTML)
- Code execution, image generation and Google search each switched on per call rather than per install
- image_save_path and only_image, for when the generated image is the output you actually want
A Gemini API key in GEMINI_API_KEY. Six further environment variables set the defaults and the ceilings: GEMINI_MODEL, GEMINI_TIMEOUT, GEMINI_MAX_OUTPUT_TOKENS, GEMINI_MAX_FILES, GEMINI_MAX_TOTAL_FILE_SIZE and GEMINI_TEMPERATURE.
One command plus a key — GEMINI_API_KEY=your_api_key npx -y aistudio-mcp-server, then supply credentials