sample-deploy s-at-scale-on-aws

This MCP server provides a centralized hub of tools and resources that can be surfaced to agents via the MCP protocol. It hosts a registry of MCP tools and exposes them through a UI (Streamlit) and API endpoints, enabling agents to discover available tools, inspect their input parameters, and trigger tool execution through the MCP workflow. The server is designed to scale across AWS deployments, leveraging common AWS services to manage registry, execution, and security. To use it, you can interact with the registry UI to explore tools and their required inputs, or integrate with a Bedrock Agent by first requesting the list of tools from the hub, selecting appropriate tools, and invoking them with the mandated input parameters. The agent’s rationale step will help determine which tool to call next based on the task, and the MCP hub will execute the chosen tool and return results for subsequent reasoning.

Prerequisites:

• Python 3.12+ (for any Python tooling or CDK components)
• Docker 24.0+ with Buildx and a running daemon
• AWS CLI configured with appropriate permissions (for deploying the stack)
• Git

Installation steps:

1. Clone the repository: git clone https://github.com/aws-samples/aws-samples-sample-deploy-mcp-servers-at-scale-on-aws.git cd aws-samples-sample-deploy-mcp-servers-at-scale-on-aws

2. Set up a Python environment and install dependencies (optional if the project uses a virtualenv or CDK in Python): python3.12 -m venv .venv source .venv/bin/activate pip install -r requirements.txt

3. Prepare AWS infrastructure (this repository provides a CDK-based deployment):

   • Ensure AWS credentials are configured
   • Review cdk.json and app.py for configuration overrides

4. Deploy the solution (CDK): ./deploy.sh

5. Access the deployed UI and endpoints:

   • Streamlit UI URL is output in the CloudFormation stack outputs under StreamlitNLBURL
   • Use the registry tab to view MCP servers and their tools

Notes:

• If you’re using Docker-based deployment locally, ensure the image name matches what is published in your registry or replace the image reference in the docker run command accordingly.

Tips and common considerations:

• The hub exposes a registry of MCP tools; use the UI to inspect required inputs before invoking tools.
• For enterprise deployments, integrate with AWS Secrets Manager and IAM roles to manage credentials and access.
• If you encounter connectivity issues between the UI and tool execution endpoints, verify VPC, security groups, and API Gateway configurations.
• Ensure Bedrock access (if using Bedrock-based agents) is configured and that the required Claude v3.5 Haiku access is provisioned in Bedrock.
• The deployment script includes destroy/shutdown commands; use them to clean up resources when tests complete.
• Custom environment variables (e.g., REGION, BEDROCK_API_KEY) may be required depending on your AWS setup and authentication model.