AWS Model Context Protocol (MCP) Server

Warnings

• gotcha The server runs as an asynchronous gRPC service. Ensure your custom `ModelContextServiceServicer` implementations are also `async def` functions and correctly handle asynchronous operations. Blocking calls within these methods can degrade performance.
  Fix: All service methods (e.g., `AddContext`, `GetContext`) overridden from `ModelContextServiceServicer` should be defined with `async def` and use `await` for any asynchronous operations.
• breaking Major changes to the underlying Model Context Protocol definition (e.g., in `mcp.proto`) could introduce breaking changes to the gRPC service interface. While `awslabs` strives for backward compatibility, new major versions of the library might reflect such protocol updates.
  Fix: Always review the changelog for breaking changes when upgrading to new major versions. Re-generating client stubs based on the updated `.proto` definition and updating your `ModelContextServiceServicer` implementation might be necessary.
• gotcha Correctly loading and managing API models and context is crucial for the server's function. The example provides a basic `AddContext`, but production implementations will require robust storage and retrieval mechanisms for the models and their associated context.
  Fix: Design a proper data store (e.g., database, S3) and logic within your `ModelContextServiceServicer` to persist and retrieve model context. Consider concurrency and consistency requirements.
• gotcha The library relies on specific versions of `grpcio` and `protobuf`. Incompatibility issues can arise if other projects in your environment use different or conflicting versions of these dependencies.
  Fix: Use a virtual environment to isolate project dependencies. If encountering issues, check `pip freeze` for `grpcio` and `protobuf` versions and align them with what `awslabs-aws-api-mcp-server` expects or is known to work with.

Install

• pip install awslabs-aws-api-mcp-server Install latest version

Imports

• McpServer

  from mcp.server import McpServer

• ModelContextServiceServicer

  from mcp.service import ModelContextServiceServicer

  The core servicer class is directly under `mcp.service`, not `mcp.server.service`.
• AddContextRequest

  from mcp.server.models.mcp_pb2 import AddContextRequest

  Generated protobuf messages are nested under `mcp.server.models`.

Quickstart

This quickstart demonstrates how to set up a basic MCP server using `awslabs-aws-api-mcp-server`. It implements a custom `ModelContextServiceServicer` and starts an asynchronous gRPC server. In a real application, you would implement the `AddContext` and `GetContext` methods to handle actual model context operations.

import asyncio
from concurrent import futures
import grpc

from mcp.server import McpServer
from mcp.service import ModelContextServiceServicer
from mcp.server.models.mcp_pb2_grpc import add_ModelContextServiceServicer_to_server
from mcp.server.models.mcp_pb2 import AddContextRequest, AddContextResponse

class MyMcpService(ModelContextServiceServicer):
    """Example implementation of the ModelContextServiceServicer."""
    async def AddContext(self, request: AddContextRequest, context: grpc.aio.ServicerContext) -> AddContextResponse:
        print(f"Received AddContext request for model: {request.model_name}")
        # Implement actual context logic here
        return AddContextResponse(success=True, message="Context added successfully")

    async def GetContext(self, request, context):
        # Implement GetContext logic
        return super().GetContext(request, context)

async def serve():
    server = grpc.aio.server(futures.ThreadPoolExecutor(max_workers=10))
    add_ModelContextServiceServicer_to_server(MyMcpService(), server)
    server.add_insecure_port('[::]:50051')
    print('Starting MCP server on port 50051')
    await server.start()
    await server.wait_for_termination()

if __name__ == '__main__':
    asyncio.run(serve())