Strategic & Tactical Context Engineering

This episode of the Naftiko Capabilities podcast features a full conversation with Patrick Kelly (SideKo, Port of Context) about context engineering for AI agents, MCP, and evolving API tooling. Patrick explains SideKo’s shift from generating developer-facing API products (docs, SDKs, mock servers from API specs, with customers like Prudential) toward Port of Context as agent consumption grows, including leveraging techniques like giving agents SDKs and dynamically exposing documentation. Kin shares his background across OpenAPI/Swagger tooling, SDK efforts (including Microsoft Kiota and dynamic SDK sizing), API management, and his Postman experience with collections and right-sizing context windows. He describesNaftiko’s work-in-progress API consumption gateway/proxy and a catalog of sandboxes for tools like Notion, GitHub, and Slack, organized into tagged capabilities to support safe experimentation for design partners (including banks) and to enable mocking and generating agent skills/cards. The conversation covers evaluating agent API usage more precisely via proxy/state control approaches, bridging enterprise governance and MCP adoption, and Port of Context’s runtime techniques (including translating many upstream MCP tools into a smaller set of agent-facing tools and exploring a virtual file system approach) while iterating safely through measurement. Patrick outlines spec linting and rules for OpenAPI-to-MCP generation, emphasizing improving descriptions so they work as tools, and handling messy real-world specs by letting agents explore context dynamically rather than stuffing everything into the prompt. Ken closes by noting the tactical vs. strategic perspectives, inviting listeners to connect for recorded conversations and sharing his contact details and recording times.