
November 23–27, 2026
Model Context Protocol servers give AI agents access to useful tools, but building every integration by hand repeats the same work across schema design, handler implementation, validation, packaging, and deployment. This session walks through a personal four-person team prototype that turns a natural-language integration request into a reviewable MCP server. The talk follows the workflow from intent capture and tool-contract planning through TypeScript/Node.js generation, protocol checks, packaging, and deployment. It explains why the system uses explicit intermediate artifacts rather than one large prompt, which checks should be deterministic, and where human approval belongs. A compact example contrasts a functionally valid but over-broad tool with one that has bounded inputs, limited authority, clear errors, isolated credentials, inspected dependencies, and malformed-request tests. Attendees will leave with a practical architecture for building agent-generated tooling without confusing fast generation with trustworthy execution. The project is personal and contains no confidential employer material.
See every session across the AI Agents Conference November 2026 line-up.
View Agenda