Prototyping in the terminal.

Claude Code

A practical guide to agentic coding from the command line for Product Design, Product Management, and Engineering teams

For: Product Management, Product Design, Data Science, and Engineering Teams

01Overview

Claude Code is Anthropic's official agentic coding tool that runs directly in your terminal. Unlike GUI-based editors, Claude Code gives you a powerful AI pair programmer that can read, write, and execute code across your entire codebase with minimal supervision.

It's designed for developers and technical teams who want the full power of Claude's coding capabilities in a familiar command-line environment, with the ability to run shell commands, manage files, and integrate with external tools through the Model Context Protocol (MCP).

02Quick Summary

Claude Code = agentic AI coding in your terminal. An AI assistant that can understand your entire codebase, execute commands, and build features end-to-end with natural language instructions.

Key Points:

  • Technology : Powered by Claude Sonnet 4 (default) or Claude Opus 4 for complex reasoning
  • Best For : Full-stack development, refactoring, debugging, code exploration, and automation
  • Interface : Terminal-based CLI with support for VS Code, JetBrains, and other editors
  • Capabilities : File operations, shell commands, web search, MCP integrations, and multi-file edits

What Makes It Different:

  • Agentic Mode : Claude can autonomously plan, execute, and iterate on tasks
  • Terminal-Native : Works in your existing workflow, no new IDE to learn
  • Full Context : Understands your entire codebase, not just open files
  • Tool Use : Can run commands, read files, search the web, and use MCP servers

03Getting Started

1. Install Claude Code

Install globally via npm:

npm install -g @anthropic-ai/claude-code

Requires Node.js 18+ and an Anthropic API key or Claude Pro/Max subscription.

2. Authenticate

Run Claude Code and follow the authentication prompts:

claude

You can authenticate with your Anthropic API key or log in with your Claude account.

3. Navigate to Your Project

Change to your project directory and start Claude Code:

cd your-project
claude

Claude will automatically index your codebase for context.

4. Start Building

Simply describe what you want to build in natural language:

> Add a dark mode toggle to the settings page

Claude will analyze your codebase, create a plan, and implement the feature.

💡 Pro Tip

Create a CLAUDE.md file in your project root with context about your codebase, coding conventions, and common patterns. Claude will automatically read this for better results.

04Tips and Tricks

Watch these walkthroughs on building your own agentic AI Chief of Staff with Claude Code:

05Core Features

Claude Code provides powerful tools for AI-assisted development directly in your terminal.

Interactive Commands

/help

Show available commands and usage

/clear

Clear conversation history and start fresh

/compact

Summarize and compress conversation context

/config

View and modify settings

/cost

Show API usage and costs for session

/doctor

Diagnose installation and configuration issues

Built-in Tools

File Operations

Read, write, edit, and search files across your codebase

Shell Commands

Execute terminal commands with your approval

Web Search

Search the web for documentation and solutions

Web Fetch

Fetch and analyze content from URLs

Git Integration

Stage, commit, and push changes with AI-generated messages

Multi-file Edits

Make coordinated changes across multiple files

Agentic Capabilities

Autonomous Execution

Claude plans and executes multi-step tasks independently

Context Awareness

Understands project structure, dependencies, and patterns

Error Recovery

Automatically handles errors and tries alternative approaches

Permission System

Asks for approval before destructive or sensitive operations

06Advanced Techniques

Power-user methods for getting the most out of Claude Code.

Effective Prompting

Be Specific

Describe exactly what you want, including edge cases and error handling

Provide Context

Reference relevant files, patterns, or documentation

Iterate Incrementally

Break large tasks into smaller, focused steps

Review and Refine

Ask Claude to explain changes and iterate on feedback

Configuration & Customization

CLAUDE.md Files

Project-specific instructions and context loaded automatically

Model Selection

Use --model flag to switch between Sonnet and Opus

Allowed Tools

Configure which commands run without asking for permission

Custom Hooks

Run scripts before/after tool executions

CLI Options

claude -p "prompt"

Run a single prompt and exit (great for scripts)

claude --continue

Resume the most recent conversation

claude --resume ID

Resume a specific conversation by ID

cat file | claude

Pipe content directly to Claude for analysis

07Claude Code for Product Managers

For years, the codebase was a black box for PMs. You could read documentation, ask engineers questions, and review PRs, but actually understanding how things worked required deep technical knowledge or patience from a willing engineer.

Claude Code changes that. You can now ask "How does our checkout flow work?" and get a real answer, with file references, logic explanations, and context about why decisions were made. You can trace a bug report through the actual code to understand root causes before triaging. You can scope features by exploring existing implementations rather than guessing at complexity.

But the real power comes from building. When you can prototype a new feature in an afternoon, you stop waiting for engineering sprints to validate ideas. You can create working demos that communicate intent better than any PRD. You can update documentation and ship it yourself with a PR.

The distance between "product idea" and "working software" shrinks dramatically, and with it, the feedback loops that make great products possible.

Example PM Session

> How is the checkout flow implemented? Walk me through the key files.

> What would it take to add Apple Pay support? Give me a rough estimate.

> Create a prototype of a simplified checkout with just credit card.

> Write a technical spec for adding Apple Pay based on how we handle Stripe.

08Claude Code for Designers

The handoff between design and engineering has always been lossy. You create a pixel-perfect mockup, write detailed specs, and then wait to see what comes back, often with subtle (or not so subtle) differences from your intent.

Claude Code lets you close that gap yourself. With the Figma MCP integration, you can import your designs directly and generate production-ready React components. But more importantly, you can iterate on the real thing: adjust that animation timing, fix the responsive breakpoint, tweak the hover state until it feels right.

This isn't about replacing engineering. It's about being able to express your intent in the actual medium. When you can prototype with real interactions, real state, and real data, you discover things that static mockups can't reveal. And when you can submit a PR with your animation fix instead of filing a ticket, everyone moves faster.

Example Designer Session

> Import my Figma frame "New Dashboard Header" and create a React component.

> Update the button hover state to have a 200ms ease-out transition.

> Show me all places where we use the Card component.

> Add a subtle entrance animation to the modal component.

09Claude Code for Data Scientists

Data science work is often bottlenecked by tooling friction. You need to write SQL to explore the data, Python to analyze it, more code to visualize it, and then figure out how to share findings with stakeholders who don't have your local environment.

Claude Code collapses these steps. You can describe the analysis you want in plain English and get a complete Jupyter notebook with exploration, visualizations, and documentation. Connect to your databases through MCP and iterate on queries conversationally. Build dbt models and data pipelines by describing the transformation logic rather than wrestling with syntax.

The real unlock is building tools for others. When creating a Streamlit dashboard or internal data app takes an afternoon instead of a sprint, you can respond to one-off requests with reusable solutions. The analysis that used to live in a notebook can become a self-service tool that empowers the whole org.

Example Data Science Session

> Create a notebook that analyzes user retention by cohort from our events table.

> Build a dbt model that calculates monthly active users with a 28-day window.

> Create a Streamlit dashboard showing key metrics from our analytics DB.

> Write a Python script to backfill missing data in the conversions table.

10Claude Code for User Researchers

Research insights are only valuable if they reach the people making decisions. But the path from raw interviews to actionable findings is paved with tedious manual work: transcription, coding, synthesis, and endless reformatting for different audiences.

Claude Code can automate the mechanical parts while you focus on the thinking. Build tools that process interview transcripts and extract themes automatically. Create searchable repositories of insights that make past research discoverable. Analyze hundreds of NPS comments or support tickets to surface patterns that would take weeks to find manually.

But perhaps the biggest unlock is prototyping for usability tests. Instead of testing static mockups or waiting for engineering to build something testable, you can create realistic prototypes with actual interactions and data. When participants can use something that feels real, you get feedback that's real too.

Example Research Session

> Build a tool that analyzes these 50 interview transcripts and extracts key themes.

> Create a prototype of the new onboarding flow for next week's usability tests.

> Analyze NPS comments from Q4 and categorize by feature area.

> Build a searchable research repository from our Notion research database.

11MCP Integrations

The Model Context Protocol (MCP) extends Claude Code with external tools and data sources.

What is MCP?

MCP is an open protocol that allows AI assistants to securely connect to external data sources and tools. With MCP servers, Claude Code can access databases, APIs, file systems, and more, extending its capabilities beyond the local codebase.

Popular MCP Servers

GitHub

Create issues, PRs, and manage repositories

Figma

Import designs and generate code from frames

PostgreSQL / SQLite

Query and modify database schemas

Slack

Send messages and retrieve channel history

Linear / Jira

Create and update issues from Claude Code

Notion

Read and write to Notion databases

Webflow

Manage sites, CMS collections, and deploy to Webflow Cloud

Setting Up MCP

Configure MCP servers in your Claude Code settings:

// ~/.claude/settings.json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "your-token"
      }
    }
  }
}

12Ship with Webflow Cloud

Once you've built your app with Claude Code, deploy it to production with Webflow Cloud, a serverless hosting platform that runs on Cloudflare's global edge network with sub-50ms response times.

The Complete Workflow

  1. Build with Claude Code : Use natural language to develop your Next.js or Astro app
  2. Push to GitHub : Claude can commit and push your changes directly
  3. Deploy on Webflow Cloud : Automatic deployments across dev, staging, and production environments

What You Get:

  • Edge-Powered Performance : Global CDN with Cloudflare Workers for lightning-fast delivery
  • Framework Support : Native Next.js and Astro support, with more frameworks coming
  • Built-in Storage : SQLite databases, Key Value Store for caching, and Object Storage for media files
  • GitHub Integration : Automated CI/CD workflows with environment-based deployments
  • Zero Infrastructure : Focus on building, not managing servers

Ready to Start Building?

Install Claude Code and start building with AI in your terminal today

Read the Claude Code Docs