Building a Dual Response MCP Server

Part 3 / 3 of "Scaling LLM-Based Reporting Past the Demo" - Theory is nice; working code is better. Walk through a complete Node.js implementation—MCP server, REST endpoints, resource store, and client-side detection—that turns natural language into paginated query results without ever bloating the LLM's context window.

December 6th, 2025

The Dual Response Pattern - Scaling LLM-Powered Reports

Part 2 / 3 of "Scaling LLM-Based Reporting Past the Demo" - MCP already has the building blocks for scalable reporting—resource_link, outputSchema, and structuredContent—you just need to know how to wire them together. This post specifies the dual response schema, REST endpoints for bulk retrieval, and the lifecycle management that keeps abandoned queries from drowning your server. November 28th, 2025

Why Your LLM Can't Be Your Reporting Layer

Part 1 / 3 of "Scaling LLM-Based Reporting Past the Demo" - Your LLM-powered reporting demo worked great—until real data turned your context window into an expensive bottleneck. Learn why asking an LLM to both generate queries and deliver results is fundamentally broken, and discover the dual-response pattern that separates these concerns. November 16th, 2025

How LLMs and Agents Will Deliver the Robot Promise

Today's robots fail in the real world because they can’t handle ambiguity, making them too expensive and limited to scale. LLMs are giving robots adaptable, language driven brains, which will get us to the tipping point where utility drives investment, cost drops, and the long-promised robot revolution begins. July 29th, 2025

Fixing SSE (MCP) Behind NGINX in a Node.js + PM2 Stack

If your MCP server is using SSE and you are behind nginx, you need to turn buffering off! July 18th, 2025

Embracing "Usually Correct"

LLMs are going to "write", in realtime, the glue between systems and tools that - until now - were too non-deterministic to be automated. July 12th, 2025

Agentic Software Architecture: Mind, Body, Tools

Software development being turned on it's head, and not just by Claude Code... The Model View Controller architecture is about to give way to the Mind Body Tools architecture - the Agentic Software Architecture. June 13th, 2025

A mental model for MCP tool development

How local MCP (Model Context Protocol) tools work with remote LLMs as a mediator with the LLM running in the cloud with locally-running tools. The key insight is that an "agent" (the local software wrapper) is different from the LLM itself, and this agent handles the stateful conversation and tool execution while using the remote LLM for natural language processing. May 16th, 2025