Aws Nova Canvas MCP

🚀 AWS Nova Canvas MCP Server

An MCP server that allows you to generate and edit images using Amazon Bedrock's Nova Canvas model.

✨ Features

• Text-to-image
• Image inpainting (de-inpainting)
• Image outpainting (extending)
• Image variants
• Image adjustment
• Color-guided generation
• Background removal
• Display image thumbnails

📦 Installation

Claude Desktop Setup

1. Clone the repository

git clone https://github.com/yunwoong/aws-nova-canvas-mcp.git

2. Configure Claude Desktop

   • Click on the menu Claude > Settings.
   • After the pop-up window appears, select Developer in the left menu, then click the Edit Settings button.
   • This will open a folder containing the settings file. The settings file is named:
   • claude_desktop_config.json

3. Add the following content to the settings file (Python version):

   • Python version

     "nova-canvas": {
       "command": "uv",
       "args": [
         "--directory",
         "Path to the cloned folder",
         "run",
         "server.py"
       ],
       "env": {
         "AWS_PROFILE": "YOUR_AWS_PROFILE"
       }
     }
     

     ✅ Only the AWS_PROFILE is required. Other variables such as AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION, and PORT are optional and not needed if your AWS configuration file is set up correctly.

     ⚙️ If the setup is successful, you can see the added "nova-canvas" project in the Claude > Settings > Developer tab. ⚠️ Important Note: The MCP settings only work on the Claude desktop application, not the web browser version.

Image Saving Location

By default, all generated or edited images will be saved in the following directories:

• macOS / Linux: ~/Desktop/aws-nova-canvas
• Windows: C:\Users\YourUsername\Desktop\aws-nova-canvas

📁 If the image saving path is not specified, the application will automatically create and use the above folders.

⚠️ Limitations

• The prompt text supports a maximum of 1024 characters.
• Each image supports a maximum of 5 model versions.
• Each image supports a maximum of 3 output formats.

📄 License

This project is licensed under the MIT License.