NODES AI 2026 - Smarter MCP Servers: Using a Graph to Solve the Context Window Problem

This talk covers the concept of smarter MCP servers, specifically focusing on using a graph to solve the context window problem. The speaker, Jonathan Giffard, compares three architectural patterns for managing tool discovery in MCP servers, highlighting the challenges of token costs associated with multiple tools. By using Neo4j Aura as an example, the session demonstrates how representing tool lists as a graph with usage weights, co-occurrence edges, and category relationships enhances the system's ability to surface relevant tools and improve over time.