Bank MCP (Managed Conversational Platform)

This project implements a Managed Conversational Platform for a bank using AWS services. The platform provides a chat interface that integrates with Claude through Amazon Bedrock and connects to the bank's API.

Architecture

The solution consists of the following components:

1. API Gateway: Provides the REST API endpoint for the chat interface
2. Lambda Functions:
   • Entry Processor: Handles incoming chat requests
   • Claude Integrator: Interacts with Claude through Amazon Bedrock
   • Bank Integrator: Connects to the bank's API
   • Response Formatter: Formats and returns the final response
3. DynamoDB: Stores conversation context and history

Configuration

The project uses environment-specific configuration files located in the config directory. The following configuration files are available:

• config/example.json: Example configuration file (safe to commit)
• config/test.json: Test environment configuration (do not commit)
• config/prod.json: Production environment configuration (do not commit)

To set up your environment:

1. Copy the example configuration:

   cp config/example.json config/test.json
   cp config/example.json config/prod.json
   

2. Edit the configuration files with your environment-specific values:

   • test.json for testing environment
   • prod.json for production environment

3. Deploy with the desired environment:

   # For test environment
   cdk deploy --context environment=test
   
   # For production environment
   cdk deploy --context environment=prod
   

Configuration Parameters

Parameter	Description	Example
environment	Environment name	"test", "prod"
bedrockModelId	Bedrock model ID	"anthropic.claude-v2"
maxTokens	Maximum tokens for Claude	"1000"
temperature	Claude temperature setting	"0.7"
bankApiUrl	Bank API endpoint	"https://api.bank.example.com"
bankApiKey	Bank API key	"your-api-key"

Setup

1. Install dependencies:

   npm install
   

2. Deploy the stack:

   cdk deploy --context environment=test
   

Usage

Send a POST request to the API Gateway endpoint with the following format:

{
  "message": "Your message here",
  "conversationId": "optional-conversation-id"
}

The response will include the conversation ID and the formatted response from the system.

Development

Each Lambda function has its own package.json file for managing dependencies. To add new dependencies to a Lambda function:

1. Navigate to the function's directory
2. Install the dependency:

   npm install <package-name>
   

Security

• The solution uses IAM roles and policies to control access to AWS services
• API Gateway is configured with appropriate authentication and authorization
• Sensitive data is stored securely in DynamoDB
• Configuration files with sensitive data are excluded from version control

Monitoring

• CloudWatch Logs are enabled for all Lambda functions
• CloudWatch Metrics are available for monitoring API Gateway and Lambda performance