In my last article, I mentioned that Claude Code was a genuine productivity multiplier while building simap-mcp. A few of you reached out asking what exactly Claude Code is and how it differs from just using Claude on the web. So here it is — a proper deep dive into the tool that’s become central to how I work.
AI-generated image via Google Nano Banana 2
Table of contents
Open Table of contents
What is Claude Code?
Claude Code is Anthropic’s official CLI for Claude. Instead of chatting in a browser window, you open your terminal, type claude, and start a conversation — right where your code lives.
That might sound like a minor difference, but it changes everything. Claude Code can:
- Read and write files directly in your project
- Run shell commands — builds, tests, git operations
- Search your codebase by file name or content
- Browse the web to look up documentation
- Edit existing code with surgical precision
It’s not just a chatbot — it’s an agent that can actually do things on your machine.
How it differs from Claude.ai
With Claude.ai, you’re in a conversation. You paste code, ask questions, copy the answers back. It works, but there’s a lot of back-and-forth. You’re the middleware between Claude and your codebase.
With Claude Code, there’s no copy-pasting. You say “fix the failing test in auth.test.ts” and it reads the file, understands the error, edits the code, and runs the test — all in one go. You approve each step, but the point is: Claude has direct access to your project.
The difference is like asking someone for directions versus having them drive you there.
The subscriptions: Pro, Max 5x, and Max 20x
Claude Code is available through Anthropic’s subscription plans:
| Plan | Price | Claude Code usage |
|---|---|---|
| Pro | $20/month | Included, basic limits |
| Max 5x | $100/month | 5x the Pro usage |
| Max 20x | $200/month | 20x the Pro usage |
Now here’s why this matters: using Claude through the API is expensive. Claude Opus costs $5 per million input tokens and $25 per million output tokens. A single heavy coding session — where Claude reads your codebase, makes changes, runs tests, iterates — can easily consume hundreds of thousands of tokens.
To put it in perspective: a few hours of intensive work with Claude Code could cost $50–100+ via the API. Do that daily, and you’re looking at $1,000–2,000/month in API bills.
With the Max 5x plan at $100/month, you get far more usage than that. The Max 20x at $200/month is essentially unlimited for most workflows. The subscription plans are dramatically cheaper than API usage for anyone who uses Claude Code seriously.
I’m on the Max 5x plan, and I’ve never come close to hitting the limit — even during heavy development weeks. For the price of what would be a day or two of API usage, I get an entire month.
Local MCP servers: extending Claude’s reach
One of Claude Code’s most powerful features is its support for MCP servers — the same protocol I wrote about in my last article. But beyond connecting to data sources, local MCP servers can give Claude eyes and hands on your entire development environment.
Chrome DevTools MCP
Take the Chrome DevTools MCP server as an example. Once configured, Claude Code can:
- Open and navigate web pages in Chrome
- Take screenshots to see what the page actually looks like
- Click elements, fill forms, type text
- Read console logs and network requests
- Run Lighthouse audits for performance analysis
Imagine you’re working on a frontend bug. Instead of describing it to Claude, you just say: “Open localhost:3000 and check if the login form works.” Claude navigates to the page, takes a screenshot, sees the issue, reads the console errors, and fixes the code — all without you switching context.
This is incredibly powerful. You go from “let me describe what I see on screen” to “Claude, look at it yourself and fix it.”
The MCP ecosystem
Chrome DevTools is just one example. There are MCP servers for databases, Supabase, GitHub, Slack, and hundreds more. Each one gives Claude a new capability. And because MCP is an open standard, anyone can build and share servers — the ecosystem is growing fast.
Built-in tools: no setup required
What surprised me when I first started using Claude Code is how much it can do out of the box, without any MCP servers at all.
File operations
- Read — view any file in your project
- Write — create new files from scratch
- Edit — make precise changes to existing files without full rewrites
- Glob — find files by pattern (e.g.,
**/*.tsx) - Grep — search file contents with regex
System operations
- Bash — run any shell command: build, test, deploy, git, you name it
Web access
- WebSearch — search the web for documentation, solutions, references
- WebFetch — fetch the contents of any URL
From the moment you install Claude Code, you have an agent that can read your code, edit it, run your tests, search the web for solutions, and commit the results — all in a single conversation. No plugins, no configuration.
When I built simap-mcp, a typical interaction looked like:
Add a new tool called
get_publication_historythat fetches the history for a given project ID. Follow the pattern of the existing tools.
Claude would:
- Read the existing tools to understand the pattern
- Read the API client to find the right endpoint
- Write the new tool following the established conventions
- Run the tests
- Fix anything that failed
Five steps, one prompt. That’s the power of having built-in tools that work together.
CLAUDE.md: giving Claude context about your project
There’s one built-in feature that deserves its own mention: CLAUDE.md. It’s a Markdown file at the root of your project that tells Claude how your codebase works — build commands, architecture decisions, conventions, anything you’d normally explain to a new team member.
You can generate one by running:
claude /initClaude will scan your project and produce a CLAUDE.md tailored to your stack. For this blog, mine includes the dev commands, the content structure, the publishing schedule, and even the branching strategy. Every time I start a new Claude Code session, it reads this file first — so it already knows how the project is organized before I ask anything.
The result: I spend less time repeating context and more time on the actual task. Instead of saying “posts go in src/data/blog/ and use this frontmatter format,” I just say “write a new post about X” and Claude already knows where to put it, what fields to include, and what conventions to follow.
If you work on a project regularly with Claude Code, CLAUDE.md is the single highest-leverage thing you can set up.
The time savings are real
I want to be honest: Claude Code doesn’t write perfect code every time. Sometimes it misunderstands the intent. Sometimes it needs a correction. But even with those iterations, the time savings are significant.
Here’s how I think about it:
- Boilerplate and scaffolding: tasks that take 30 minutes manually take 2 minutes with Claude Code. New files, configs, test setups — it handles the repetitive parts.
- Debugging: instead of manually tracing through code, Claude reads the error, traces the call stack, and suggests (or applies) the fix. What used to be an hour of
console.logdebugging is now a 5-minute conversation. - Learning new codebases: Claude can read and explain code faster than I can skim it. When I jump into an unfamiliar repo, I start with “explain how the auth flow works” and get a structured walkthrough.
- Refactoring: rename a function across 20 files? Change an API response shape and update every consumer? Claude does this reliably and fast.
The real multiplier isn’t that Claude writes code faster — it’s that it removes context-switching. I stay in the terminal, in the flow, thinking about what I want to build rather than how to type it out.
Claude Code on the web
There’s one more thing worth mentioning: Claude Code is also available on the web at claude.ai. When you connect a GitHub repository, Claude can work on your code directly in the browser — no terminal needed.
The workflow is straightforward:
- Open a conversation on claude.ai
- Connect your GitHub repository
- Describe what you want changed
- Claude makes the changes across your codebase
- Review the diff and create a pull request in one click
This is perfect for when you’re away from your development machine, or for tasks that don’t need heavy local testing. I’ve used it to fix typos, update documentation, and even implement small features — all from my phone.
The pull request integration is particularly smooth. Claude generates a proper PR with a clear title and description of the changes. Your CI pipeline runs, reviewers get notified — it fits right into a normal development workflow.
Wrapping up
Claude Code has genuinely changed how I approach development work. It’s not a replacement for understanding your code — you still need to review what it produces and guide it in the right direction. But as a tool that handles the mechanical parts of coding while you focus on the design and architecture? It’s the best I’ve used.
If you’re already using Claude and haven’t tried Claude Code, I’d recommend giving it a shot. Start with the Pro plan, open your terminal, and type claude. You might be surprised how quickly it becomes part of your workflow.
Thanks for reading! If you have questions or want to share your own experience with Claude Code, feel free to reach out via email or LinkedIn.
Next up, I’ll be exploring BMAD — an open-source method for structuring AI-assisted development from project vision to implementation. If you’re curious, check out the BMAD Method repository in the meantime. See you in two weeks!