{
  "id": 17083259,
  "title": "Why MCP servers should tell agents what to do next",
  "url": "https://urgent.news/2026/10/29/why-mcp-servers-should-tell-agents-what-to-do-next",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-10-29T15:55:39.000Z",
  "source": {
    "name": "The New Stack",
    "slug": "the-new-stack",
    "url": "https://thenewstack.io/aws-mcp-agent-observability/"
  },
  "original_language": "en",
  "account": "James Ward, a principal developer advocate at AWS, argues that many MCP servers currently function as simple APIs, retrieving data upon request. However, Ward envisions MCP servers as more than just data providers; he sees them as integral members of the conversation, much like a third or fourth party. To illustrate his point, Ward highlighted AWS's approach to MCP servers, which he believes can be enhanced by providing more context and visibility into the agent's behavior.\n\nCurrently, MCP servers lack the ability to attach a trace ID to tool responses, making it challenging for developers to track the entire conversation. AWS addresses this limitation by including a session ID parameter in the tool responses, allowing agents to pass it along in subsequent calls. Additionally, a second parameter can carry the user's prompt or a summary of their goal. By doing so, teams can compare the user's original request with the agent's action sequence, evaluate the performance, and refine tool descriptions or parameters accordingly.\n\nAWS has already implemented these enhancements for its own MCP servers. In May, AWS opened the MCP Server to the public, and in September, AWS added Lambda diagnostics to the servers. Ward shared an example from the Agentic AI Foundation profile, where an agent fetched customer records one at a time due to the server's limited toolset. By observing the full conversation, teams could identify this pattern and develop a tool that retrieves multiple records at once.\n\nWard emphasized that teams should not limit themselves to MCP as the sole solution. AWS also provides a CLI and skills for agents, acknowledging that the OAuth model of MCP may not be as flexible as the CLI for managing multiple AWS user profiles. AWS conducts evaluations to compare different approaches, ensuring that teams can choose the best fit for their specific needs.\n\nIn summary, Ward's vision for MCP servers goes beyond simply retrieving data. He advocates for MCP servers to act as active participants in the conversation, providing guidance, visibility, and the ability to measure and improve the overall agent-human collaboration. By observing the complete journey, teams can identify areas for improvement, build better tools, and ultimately enhance the performance of both agents and humans.",
  "summary": "Many MCP servers do one thing. An agent asks for data, and the server hands it back. AWS’s James Ward The post Why MCP servers should tell agents what to do next appeared first on The New Stack .",
  "key_points": [
    "MCP servers currently act as simple APIs, retrieving data upon request.",
    "AWS plans to enhance MCP servers by providing more context and visibility into agent behavior."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}