Loading video player...
Documentation used to eat 100K tokens of my context window. Now it uses 2,800. Two MCP servers solved the context rot problem: ref.tools and exa.ai. I tested them across Claude Code, Cursor, Codex, and Droid Factory on a real Tailwind v4 refactoring task. The results: Claude Code used 2,800 tokens (1.4% of context). In this video, I walk through the exact setup process for all four tools and show you the one prompting technique that makes these MCPs actually work. No theory, just the configuration files and terminal commands. GET THE TOOLS - ref.tools MCP: https://rfer.me/reftools - exa.ai MCP: https://rfer.me/exaai - Join my Discord: https://rfer.me/discord - My LLM Rules & Prompts Repo: https://rfer.me/rules š Key moments 00:00 Introducing ref.tools and exa.ai MCPs 02:08 Configuring ref.tools in Claude Code 03:27 Configuring exa.ai in Claude Code 04:37 Demo: Using MCPs to upgrade a Tailwind CSS project 07:22 How to configure MCPs in Cursor 10:14 Comparing Claude Code vs. Cursor performance 12:37 Setting up MCPs for Codex 15:41 Setting up MCPs for DroidFactory 16:40 Pro-tip: How to prompt agents to use MCPs 18:00 The future of AI coding and final thoughts CONNECT WITH RAY X (Twitter): https://x.com/RayFernando1337 Weekly AI Insider Newsletter: https://dub.sh/RayMasterAI #MCPservers #ClaudeCode #Cursor #AIcoding #ContextWindow #SoftwareEngineering #RayFernando #Tailwindv4 #AgenticCoding