Designing MCP tools for agents: Lessons from building Datadog's MCP server

Reilly Wood discusses the evolution and design improvements of Datadog's Model Context Protocol (MCP) server, which is tailored for AI agents to enhance observability. Initially, the server was a simple API wrapper, but real-world usage revealed inefficiencies, such as context window overloads and resource mismanagement. To address these, Datadog optimized data formats by using CSV and YAML over JSON, implemented token-aware pagination, and enabled SQL querying for more efficient data handling. These changes reduced context space usage and costs, while improving accuracy. Wood also highlights the importance of guiding agents with specific error messages and accessible documentation, and outlines approaches like flexible tools and layering to manage tool count and complexity. The article contrasts the general-purpose MCP server with specialized agents like Bits AI SRE, emphasizing the trade-offs between flexibility and workflow-specific optimizations. The insights gained from these developments are shaping Datadog's approach to building scalable, agent-friendly systems, with an eye on future advancements in the field.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.