Session State Management

The Lambda MCP Server includes built-in session state management that persists across tool invocations within the same conversation. This is particularly useful for tools that need to maintain context or share data between calls.

Session data is stored in a DynamoDB table against a sessionId key. This is all managed for you.

Here's an example of how to use session state:

from lambda_mcp.lambda_mcp import LambdaMCPServer

session_table = os.environ.get('MCP_SESSION_TABLE', 'mcp_sessions')

mcp_server = LambdaMCPServer(name="mcp-lambda-server", version="1.0.0", session_store=session_table)

@mcp_server.tool()
def increment_counter() -> int:
    """Increment a session-based counter."""
    # Get the current counter value from session state, default to 0 if not set
    counter = mcp_server.session.get('counter', 0)
    
    # Increment the counter
    counter += 1
    
    # Store the new value in session state
    mcp_server.session['counter'] = counter
    
    return counter

@mcp_server.tool()
def get_counter() -> int:
    """Get the current counter value."""
    return mcp_server.session.get('counter', 0)

The session state is automatically managed per conversation and persists across multiple tool invocations. This allows you to maintain stateful information without needing additional external storage, while still keeping your Lambda function stateless.

Authentication

The sample server stack uses Bearer token authentication via an Authorization header, which is compliant with the MCP standard. This provides a basic level of security for your MCP server endpoints. Here's what you need to know:

1. Bearer Token: When you deploy the stack, a bearer token is configured through a custom authorizer in API Gateway

2. Using the Bearer Token:

   • The client must include the bearer token in requests using the Authorization header with the format: Bearer <your-token>
   • The token value is provided in the stack outputs after deployment
   • The sample client is configured to automatically include this header when provided with the token

3. Custom Authorizer: The implementation uses a simple custom authorizer that validates a single bearer token. This can be easily extended or replaced with more sophisticated authentication systems like Amazon Cognito for production use.

⚠️ Security Note: While bearer token authentication provides a standard-compliant authentication mechanism, consider implementing additional security measures such as:

• AWS IAM roles and policies
• OAuth 2.0 / JWT with proper token management
• Amazon Cognito User Pools

The current bearer token implementation is primarily intended for demonstration and development purposes. For production systems handling sensitive data, implement appropriate additional security measures based on your specific requirements.

What is this all about?

This is a proof-of-concept implementation of an MCP server running on AWS Lambda, along with a Strands Agents client that demonstrates its functionality. The project consists of two main components:

1. Lambda MCP Server: A Python-based serverless implementation that makes it incredibly simple to deploy cloud hosted MCP tools.
2. Strands Agents Client: A demonstration client that shows how to interact with the Lambda MCP server using Strands Agents.

Example Tools

The server comes with three example tools that demonstrate different use cases:

1. get_weather(city: str): Get the current weather for a city. Returns a simulated temperature for the given city.
2. count_s3_buckets(): Count the number of S3 buckets in your AWS account.
3. get_time(): Get the current UTC date and time, and show how long since last time was asked for (if session available).

Getting Started

Prerequisites

• AWS Account with appropriate permissions
• AWS SAM CLI installed
• Access to Amazon Bedrock in your AWS account
• Amazon Nova Pro enabled in your Amazon Bedrock model access settings

Amazon Bedrock Setup

Before running the client, ensure you have:

1. Enabled Amazon Bedrock access in your AWS account
2. Enabled the Amazon Nova Models in your Bedrock model access settings
3. Appropriate IAM permissions to invoke Bedrock APIs

Server Deployment

1. Clone this repository:

   git clone <repository-url>
   

2. Navigate to the server directory:

   cd server-http-python-lambda
   

3. Deploy using SAM:

   sam build && sam deploy --guided
   

   Note: You will be prompted for an McpAuthToken. This is the Authorization Bearer token that will be requitred to call the endpoint. This simple implimentation uses an AWS API Gateway authorizers with the McpAuthToken passed in as an env var. This can be swapped out for a production implimentation as required.

Client Setup

1. Update the api_gateway_url and auth_token in ./client-strands-agents-mcp/main.py:

   cd client-strands-agents-mcp
   

2. Navigate to the client directory:

   cd client-strands-agents-mcp
   

3. Run the helper script:

   uv sync
   uv run main.py
   

Adding New Tools

The Lambda MCP Server is designed to make tool creation as simple as possible. Here's how to add a new tool:

1. Open server/app.py
2. Add your new tool using the decorator pattern:

@mcp_server.tool()
def my_new_tool(param1: str, param2: int) -> str:
    """Your tool description.
    
    Args:
        param1: Description of param1
        param2: Description of param2
        
    Returns:
        Description of return value
    """
    # Your tool implementation
    return f"Processed {param1} with value {param2}"

That's it! The decorator handles:

• Type validation
• Request parsing
• Response formatting
• Error handling
• MCP Documentation generation

Notes

• For production use, consider adding authentication and authorization using AWS IAM best practices

Security

See CONTRIBUTING for more information.

For AWS security best practices, refer to the AWS Security Documentation and Amazon Bedrock security best practices.

License

This library is licensed under the MIT-0 License. See the LICENSE file.

Changes

Version 2.0.0

• The Lambda MCP Server module code is no longer included