Descriptions:
Edmund Yong walks through the development of new features for CreateSkills, a browser extension and custom MCP server he’s building that saves online content as clean markdown for use by AI agents. The core problem he’s solving is library clutter — as the number of saved sources grows, finding relevant content manually becomes impractical — so he adds two capabilities: automatic source organization (grouping related items into folders) and semantic source retrieval (ranking saved content by relevance to a task).
The standout tool introduced in the video is Ghost, a free service that lets AI coding agents spin up, fork, query, and discard their own PostgreSQL databases autonomously via an MCP server. Yong installs Ghost via Homebrew, connects it to Codex using the `ghost mcp install` command, and uses it to test new features against a cloned copy of his production data — avoiding any risk of corrupting real records. He notes Ghost offers 100 hours of compute and 1 TB of storage on its free tier with hard spending caps enabled by default.
The second half of the video covers Yong’s updated AI coding stack, explaining how his tooling has shifted significantly over the past few months. The overall workflow demonstrates how MCP-connected agents can handle the full build-test-iterate loop for backend features without manual database management, which he frames as the next step beyond simple code generation.
📺 Source: Edmund Yong · Published May 27, 2026
🏷️ Format: Hands On Build
