MCP 服务器

AWS Nova Canvas MCP Server

一个 MCP 服务器，允许您使用 Amazon Bedrock 的 Nova Canvas 模型来生成和编辑图像，支持诸如文本到图像生成、图像修复（inpainting）、图像外延（outpainting）、图像变体和背景移除等功能。

Tools

text_to_image

Generate an image from a text prompt. After generation, you can use the show_image tool to view the thumbnail. Args: prompt: Text prompt for generating an image (maximum 1024 characters) negative_prompt: Text prompt for excluding attributes from generation (maximum 1024 characters) height: Image height (pixels) width: Image width (pixels) num_images: Number of images to generate (maximum 4) cfg_scale: Image matching degree for the prompt (1-20) seed: Seed value for image generation open_browser: Whether to open the image in the browser after generation Returns: Dict: Dictionary containing the file path of the generated image and the thumbnail image

inpainting

Inpaint a specific part of an image using a text mask prompt. Args: image_path: File path of the original image prompt: Text prompt for the area to be inpainted mask_prompt: Text prompt for specifying the area to be masked (e.g., "window", "car") negative_prompt: Text prompt for excluding attributes from generation height: Output image height (pixels) width: Output image width (pixels) cfg_scale: Image matching degree for the prompt (1-20) open_browser: Whether to open the image in the browser after generation Returns: Dict: Dictionary containing the file path of the inpainted image

outpainting

Expand the image to create an outpainting. Args: image_path: File path of the original image mask_image_path: File path of the mask image prompt: Text describing the content to be generated in the outpainting area negative_prompt: Text specifying attributes to exclude from generation outpainting_mode: Outpainting mode (DEFAULT or PRECISE) height: Output image height (pixels) width: Output image width (pixels) cfg_scale: Prompt matching degree (1-20) Returns: Dict: Dictionary containing the file path of the outpainted image

image_variation

Generate a new variation of the input image while maintaining its content. Args: image_paths: List of file paths of the original images (1-5) prompt: Text for generating a variation image (optional) negative_prompt: Text specifying attributes to exclude from generation similarity_strength: Similarity between the original image and the generated image (0.2-1.0) height: Output image height (pixels) width: Output image width (pixels) cfg_scale: Prompt matching degree (1-20) Returns: Dict: Dictionary containing the file path of the variation image

image_conditioning

Generate an image that follows the layout and composition of a reference image. Args: image_path: File path of the reference image prompt: Text describing the image to be generated negative_prompt: Text specifying attributes to exclude from generation control_mode: Control mode (CANNY_EDGE, etc.) height: Output image height (pixels) width: Output image width (pixels) cfg_scale: Prompt matching degree (1-20) Returns: Dict: Dictionary containing the file path of the generated image

color_guided_generation

Generate an image using a specified color palette. Args: prompt: Text describing the image to be generated colors: List of color codes (1-10 hex color codes, e.g., "#ff8080") reference_image_path: File path of the reference image (optional) negative_prompt: Text specifying attributes to exclude from generation height: Output image height (pixels) width: Output image width (pixels) cfg_scale: Prompt matching degree (1-20) ctx: MCP context Returns: Dict: Dictionary containing the file path of the generated image

background_removal

Remove the background of an image automatically. Args: image_path: File path of the original image ctx: MCP context Returns: Dict: Dictionary containing the file path of the image with the background removed

show_image

Create a thumbnail of the image and return it. The maximum size is 1048578. Supports URLs or local file paths. Args: image_path: Image URL or local file path width: Output image width (pixels) height: Output image height (pixels) Returns: Image: Thumbnail image

README

<h2 align="center"> AWS Nova Canvas MCP 服务器 </h2>

一个 MCP 服务器，允许您使用 Amazon Bedrock 的 Nova Canvas 模型生成和编辑图像。

特性

文本到图像
图像修复
图像外绘
图像变体
图像条件控制
颜色引导生成
背景移除
显示图像缩略图

安装

Claude 桌面设置

克隆仓库

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

配置 Claude 桌面
- 从 Claude 桌面菜单中点击 Claude > Settings。
- 当弹出窗口出现时，从左侧菜单中选择 Developer，然后点击 Edit Settings 按钮。
- 这将打开一个包含设置文件的文件夹。此设置文件的名称是：
- claude_desktop_config.json

将以下内容添加到设置文件中（Python 版本）：
- python 版本
```
"nova-canvas": {
  "command": "uv",
  "args": [
    "--directory",
    "克隆文件夹的路径",
    "run",
    "server.py"
  ],
  "env": {
    "AWS_PROFILE": "您的_AWS_PROFILE"
  }
}
```
  ✅ 只需要 AWS_PROFILE。如果您的 AWS 配置文件设置正确，则 AWS_ACCESS_KEY_ID、AWS_SECRET_ACCESS_KEY、AWS_REGION 和 PORT 等其他变量是可选的，不需要。
  
  ⚙️ 如果设置成功完成，您可以在 Claude > Settings > Developer tab 中看到已添加 "nova-canvas" 项。 ⚠️ 重要提示： MCP 设置仅适用于 Claude 桌面应用程序，不适用于 Claude Web 浏览器版本