The Model Context Protocol (MCP) is the fastest growing standard in AI tooling. In 6 months it grew 232%, with 10,000+ public servers now available for Claude, Cursor, Windsurf, and VS Code. Yet the biggest bottleneck to adoption remains discovery — developers don't know which servers exist, how to install them, or how to govern them at team level.

This talk tells the story of building MCPNest — the marketplace for MCP servers — and what we learned from indexing 7,561+ servers in two weeks.

Part 1: The Discovery Problem

Why finding MCP servers is still broken. Five different install_config formats across thousands of servers. How we built a quality scoring system (A to F) and what separates a great server from a broken one.

Part 2: The Recursive Loop

The most interesting architectural decision: MCPNest is both the registry and an MCP server. Install mcpnest-mcp once, and you can search 7,561+ servers using natural language inside Claude or Cursor. npm lives inside Node. Docker Hub lives inside Docker. MCPNest lives inside Claude.

Part 3: The Enterprise Problem

Individual developers self-serve. But enterprise teams face a different challenge: which servers has IT approved? Which version is the team running? Who installed what, and when? We built an Enterprise workspace with private registry, version control, health monitoring, and audit log — the same governance layer that enterprise teams need before they can adopt MCP at scale.

Part 4: What's next — the Gateway

One URL per workspace. All MCP servers behind it. Auth, logging, and routing centralized. Launching June 2026. Why this matters for enterprise AI adoption.

The talk is practical and demo-heavy. Attendees will see MCPNest in action and leave with a clear mental model of how MCP discovery, installation, and governance work — and how Portugal fits into this global ecosystem.