Why MCP servers should tell agents what to do next

Many Model Context Protocol (MCP) servers function as basic APIs, returning data without guiding AI agents on what to do next. AWS Principal Developer Advocate James Ward believes MCP servers should play a more active role in agent workflows, providing context, recommendations and actionable next steps.

Speaking at AGNTCon + MCPCon in San Jose, Ward explained how better observability and evaluation can improve agent performance. Because MCP lacks built-in trace IDs, AWS uses session identifiers to track tool calls across workflows, helping developers understand how agents interpret requests, identify inefficiencies and complete tasks.

By analyzing these interactions, teams can refine tool descriptions, introduce more efficient capabilities and measure improvements through evaluations. Ward also emphasized that MCP isn't always the best integration approach, noting that AWS evaluates alternatives such as command-line interfaces and skills.

Ultimately, Ward advocates applying established software engineering practices to agentic AI: observe workflows, identify problems, test improvements and continuously optimize how humans and AI agents collaborate to accomplish tasks.

Learn more from The New Stack around the latest developments in rapid evolution of developer and agent experiences:

The 3 roles AI agents play in your developer platform

Your agent context needs a development lifecycle 

AWS open-sources an AI agent it says is 45% cheaper than Claude Code and Codex 

Join our community of newsletter subscribers to stay on top of the news and at the top of your game.