Skip to content

Latest commit

 

History

History
128 lines (94 loc) · 4.34 KB

README.md

File metadata and controls

128 lines (94 loc) · 4.34 KB
Raw

Amazon Bedrock MCP Server

A Model Control Protocol (MCP) server that integrates with Amazon Bedrock's Nova Canvas model for AI image generation.

Amazon Bedrock Server MCP server

Features

  • High-quality image generation from text descriptions using Amazon's Nova Canvas model
  • Advanced control through negative prompts to refine image composition
  • Flexible configuration options for image dimensions and quality
  • Deterministic image generation with seed control
  • Robust input validation and error handling

Prerequisites

  1. Active AWS account with Amazon Bedrock and Nova Canvas model access
  2. Properly configured AWS credentials with required permissions
  3. Node.js version 18 or later

Installation

AWS Credentials Configuration

The server requires AWS credentials with appropriate Amazon Bedrock permissions. Configure these using one of the following methods:

  1. Environment variables:

    export AWS_ACCESS_KEY_ID=your_access_key
export AWS_SECRET_ACCESS_KEY=your_secret_key
export AWS_REGION=us-east-1  # or your preferred region

  2. AWS credentials file (~/.aws/credentials):

    [the_profile_name]
aws_access_key_id = your_access_key
aws_secret_access_key = your_secret_key

    Environment variable for active profile:

    export AWS_PROFILE=the_profile_name

  3. IAM role (when deployed on AWS infrastructure)

Claude Desktop Integration

To integrate with Claude Desktop, add the following configuration to your settings file:

MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "amazon-bedrock": {
      "command": "npx",
      "args": [
        "-y",
        "@zxkane/mcp-server-amazon-bedrock"
      ],
      "env": {
        "AWS_PROFILE": "your_profile_name",         // Optional, only if you want to use a specific profile
        "AWS_ACCESS_KEY_ID": "your_access_key",     // Optional if using AWS credentials file or IAM role
        "AWS_SECRET_ACCESS_KEY": "your_secret_key", // Optional if using AWS credentials file or IAM role
        "AWS_REGION": "us-east-1"                   // Optional, defaults to 'us-east-1'
      }
    }
  }
}

Available Tools

generate_image

Creates images from text descriptions using Amazon Bedrock's Nova Canvas model.

Parameters

  • prompt (required): Descriptive text for the desired image (1-1024 characters)
  • negativePrompt (optional): Elements to exclude from the image (1-1024 characters)
  • width (optional): Image width in pixels (default: 1024)
  • height (optional): Image height in pixels (default: 1024)
  • quality (optional): Image quality level - "standard" or "premium" (default: "standard")
  • cfg_scale (optional): Prompt adherence strength (1.1-10, default: 6.5)
  • seed (optional): Generation seed for reproducibility (0-858993459, default: 12)
  • numberOfImages (optional): Batch size for generation (1-5, default: 1)

Example Implementation

const result = await callTool('generate_image', {
  prompt: "A serene mountain landscape at sunset",
  negativePrompt: "people, buildings, vehicles",
  quality: "premium",
  cfg_scale: 8,
  numberOfImages: 2
});

Prompt Guidelines

For optimal results, avoid negative phrasing ("no", "not", "without") in the main prompt. Instead, move these elements to the negativePrompt parameter. For example, rather than using "a landscape without buildings" in the prompt, use "buildings" in the negativePrompt.

For detailed usage guidelines, refer to the Nova Canvas documentation.

Development

To set up and run the server in a local environment:

git clone https://github.com/zxkane/mcp-server-amazon-bedrock.git
cd mcp-server-amazon-bedrock
npm install
npm run build

Performance Considerations

Generation time is influenced by resolution (width and height), numberOfImages, and quality settings. When using higher values, be mindful of potential timeout implications in your implementation.

License

This project is licensed under the MIT License - see the LICENSE file for details.