Lesson 06 · Claude Mastery Pro+ ~16 min read Setup + 3 workflows

Claude Code: the CLI for real work.

Claude Code is Anthropic's command-line tool that gives Claude direct access to your codebase, terminal, and file system. It's the most powerful AI coding tool available — and the most dangerous if misused. This lesson covers setup, the prompt patterns that ship code, MCP servers, and the workflows that justify the Pro+ tier.

The mental model

Claude Code isn't an IDE assistant. It's a junior engineer who can run commands.

The mental model is delegating to a junior engineer who can read your code, write code, run tests, search the web, and make commits. You're the senior who reviews and approves. The difference between this and Copilot/Cursor is autonomy: Claude Code does multi-step work and then reports back.

Workflow 01 Setup and first task

1

Install, authenticate, run your first task

Install Claude Code (`npm install -g @anthropic-ai/claude-code` or via your platform's package manager). Run `claude` in any repo. Start with low-stakes tasks.

The prompt that works

First-task patternclaude > Add JSDoc comments to every public function in src/utils/. Run the existing tests after to make sure nothing broke. Show me a diff of what you changed before committing.

Best use cases

  • Documentation generation
  • Test coverage additions
  • Repetitive refactors with low blast radius
  • Codebase Q&A ("how does authentication work in this repo?")
Always start with `--dry-run` mode on unfamiliar codebases until you trust the patterns. Never let Claude Code commit and push in one shot; review the diff first.
Time savings: Setup takes 10 min. First useful task: immediately.

Workflow 02 The MCP server pattern

2

Connect Claude Code to your real tools

MCP (Model Context Protocol) servers let Claude Code access your tools — your Linear, your GitHub, your Slack, your database. Configured once, available forever.

The prompt that works

Setup patternInstall MCP servers for: GitHub (PR review), Linear (issue context), Postgres (database queries). Now Claude Code can reference issues, query databases, and read PR diffs as part of its work.

Best use cases

  • Issue → code → PR workflow within Claude Code
  • Database-aware code (it can query the schema)
  • PR review with full context (it sees the description, comments, related issues)
  • Production-aware debugging (with logs/observability MCP servers)
MCP server permissions matter. Don't connect a production database with write access; use read-only credentials. Same for any sensitive system.
Time savings: Context-switching between tools: massive reduction.

Workflow 03 Multi-step development sessions

3

Real feature development end-to-end

Give Claude Code a feature brief. It plans, codes, tests, and prepares the PR. You review at key checkpoints.

The prompt that works

Feature brief exampleImplement feature: API endpoint to soft-delete user accounts. Requirements: PATCH /users/:id/delete sets deleted_at, returns 204. Update User model. Add migration. Write unit tests. Update API docs. Don't change existing endpoints. Plan first, show me, then execute step by step. Ask before destructive operations.

Best use cases

  • New feature development with clear specs
  • Bug fixes with reproducible test cases
  • Refactors across multiple files
  • Documentation generation tied to code changes
Always require the plan-first pattern. Letting Claude Code charge ahead on complex features without a plan checkpoint leads to scope drift and wrong solutions.
Time savings: Feature dev time: 30-50% reduction with proper supervision.

Final challenge: ship something with Claude Code

Pick a small feature or chore in your current project. Use Claude Code end-to-end: brief, plan, code, test, PR. Review carefully. Notice which parts saved time vs. which needed your full attention. That calibration is the actual skill.

What you can do now

  • Install Claude Code and authenticate it for your accounts
  • Configure MCP servers for the tools you use daily
  • Use the plan-first pattern for any multi-step work
  • Recognize the boundaries (when to delegate vs. when to drive yourself)
  • Maintain safe-guard patterns (no auto-commit, read-only DB access, etc.)
Pro
Up next in Claude Mastery

Lesson 7 · Custom instructions: teach Claude your style once

How to write custom instructions that genuinely change output. The 4 sections every good instruction set needs, and the anti-patterns that make Claude worse. See pricing →