MCP NOW CONFERENCE

The "ugly" truth about today's MCP tools? They're noisy, insecure, unreliable and demand tedious manual configuration. We encountered the same frustrations with existing open-source solutions. Our answer? We built our own. Come hear the technical journey behind our working full-stack solution for managed MCP servers and connectivity.

Over the past few months, I've explored how Model Control Protocols (MCPs) can transform Cursor from a helpful assistant into a true coding copilot. By integrating custom MCP servers and crafting focused cursorrule files, I've automated the tedious parts of my workflow—like writing detailed PR messages—and boosted the parts I actually care about. This talk shares what worked, what didn’t, and how to build your own smarter, more opinionated Cursor experience.

We built a conversational analytics system where a central LLM interfaces with the user and uses MCP to orchestrate a network of specialized, tool-augmented agents. The goal: make organizational data accessible to anyone through natural language — no SQL, schema knowledge, or dashboards required.

The central LLM routes tasks to agents dedicated to specific roles: querying databases, running deep data investigations, visualizing results, and — crucially — learning over time. Our Memory Agent observes all interactions and query runs, stores insights, and improves future performance by adapting responses and optimizing tool usage. Together, these agents operate as an intelligent, modular system controlled via MCP.

In this talk, we’ll share the architecture, development workflow, and practical tips for building and deploying multi-agent systems with MCP. We’ll cover lessons from experimentation, how to design reusable agents, and how to enable memory and learning across agent runs.

Present our sterile environment(s) for AI agents to create components at scale, then integrate them with our agentic IDE.

In this lecture, I'll explain the road we took from basic MCP to hosted MCP. How do we make it work with different clients like Claud, Copilot, Cursor… I also deep dive into how we overcome the limit tool and the dynamic tool usages we made (we are using graphql for api calls).

How ElevenLabs is creating personal voice assistants with MCP + Conversational AI