Building an MCP server to expose agent tools via Strands Agents

This second part of the Once Upon Agentic AI series shows how to move a tool out of an agent's process and into a separate Model Context Protocol server. It builds on the first part, where a Strands agent ran a dice-rolling function locally, and keeps the same agent loop while changing only where the function executes. The article explains MCP's client-server architecture, in which a host creates a client per server, and lists the three server primitives: prompts, resources, and tools. It also covers the JSON-RPC 2.0 message format and the two available transport mechanisms for carrying those messages. Setup requires Python, uv, and AWS credentials with access to Amazon Bedrock, and the accompanying code is available in an AWS samples repository.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in