Because the enable flags are independent, you can start with the one account you already have and add the others later, and list_models reports what is actually reachable rather than what the documentation advertises. The repository's own example runs a single architecture diagram through three providers side by side.
An MCP server that puts several image-generation providers behind one set of tools: Amazon Nova Canvas on Bedrock, OpenAI's GPT Image and DALL-E, and Google's Gemini and Imagen models.
- The models currently reachable across every provider you have enabled, in one list (list_models)
- An image generated from a text prompt on a model you name (generate_image)
- An existing image on disk transformed according to a prompt (transform_image)
- Prompt-writing guidance and worked examples, so the prompt is not the weak link (prompt_guide)
- Providers enabled independently, so the server runs with only the accounts you have
A recent Python, and at least one provider turned on: ENABLE_AWS with AWS_PROFILE and AWS_REGION, ENABLE_OPENAI with OPENAI_API_KEY, or ENABLE_GEMINI with GEMINI_API_KEY.
One command plus a key — uvx universal-image-mcp@latest, then supply credentials