Claude Code setup: MCP, skills, subagents and memory

This intermediate Claude Code tutorial assumes viewers already know how to install and use the coding assistant. The speaker explains model selection and commands for usage reports, context inspection and conversation compaction. He recommends disabling unused MCP tools to reduce context overhead.

The setup walkthrough covers GitHub MCP authentication with a personal access token and distinguishes local, project and user scopes. In the repository demo, Claude initially tries the unavailable GH command before the speaker redirects it to the MCP server. Context7 supplies documentation access, while Playwright supports browser testing. The speaker says the Superpowers plugin adds workflows but increases context use and slows his sessions.

A custom code review skill turns repeated instructions into a Markdown workflow. The demo requires a session restart before the new slash command appears. The tutorial then creates a test-writing subagent with a selected model, tool permissions and project memory, explaining how an AI agent with its own context differs from a reusable skill.

The sponsored Nimbalyst segment demonstrates a session Kanban board and rendered Markdown diffs; the speaker describes it as open source. The final section demonstrates CLAUDE.md for carrying written information into new sessions and discusses a larger folder structure for project memory.