DEV Community: Neo Zino

How to make Claude Code more than a coding agent: Google traffic while you sleep (open-source) 🤖📈

Neo Zino — Mon, 10 Aug 2026 09:07:11 +0000

On August 7 Anthropic announced that auto mode becomes the default in Claude Code. From August 14, sessions on Pro, Max and Team stop asking permission for every command. Anthropic's own number for what that changes: Claude works 9x longer between interruptions.

Anthropic's own announcement. From August 14 your agent stops asking permission.

I decided to make my Claude Code an SEO manager. It's excellent at finding keywords, researching, and drafting articles and interactive tools, if you give it the right access.

Google traffic compounds: one guide that ranks keeps sending people every month, for free, while you sleep.

That's the job I gave mine. Everything below is open source, so you can copy the whole thing completely for free.

The advantage

There's an advantage here nobody uses. Every SEO tool on the market learns about your product by crawling your homepage. Your agent doesn't have to. It already knows your product, because it probably built it. It has the codebase, the README, the real features.

It knows things your homepage doesn't even mention.

So instead of paying $99/month for a tool to badly reverse-engineer what your agent already knows, give the agent the job.

1. Hand it the job description

DispatchSEO (mine, and open source) is a fully self-hostable tool that lets your agent run your SEO: research keywords, check Google rankings, draft content, identify trends and AI citations.

What you need:

Your site's source in a GitHub repo. Content ships as pull requests, so git-based sites only. Sorry WordPress.
A machine with Docker and about 1 GB of RAM.
Google Search Console access.
Claude Code, on the subscription you already pay for.

Then:

docker compose up. Database, migrations, schedules and the headless builder are all in the box.
Run the setup wizard: connect Search Console and your repo.
Paste one line into Claude Code. Your agent now has 61 tools over MCP: the queue, keywords, rankings, Search Console data, backlink prospects, trend scanning.
Turn the schedule on.

The dashboard. Search Console is free, and your own impression data beats any seed list.

One honest note about "while you sleep": schedules only run while the machine is awake. A laptop is fine for a test drive. A $5 VPS or an old desktop that never sleeps is the real move.

On Codex or Cursor instead? Both work, overnight builder included. The backend doesn't care which agent turns up for the shift.

2. It researches from your own codebase

Every run, the agent re-reads the repo and derives keyword candidates from the product itself. It hunts commercial intent first: "X vs Y", "best X for Y", "X alternatives". Easy informational traffic converts at zero, so it isn't the target.

Survivors get validated for volume and difficulty, then the agent reads Google's actual page 1 to see whether the keyword is winnable. Everything lands in a queue with the reasoning attached, so you can see why it wants to write something before it writes it.

The queue, with the reasoning attached. Approve, reject, reorder, or set it to auto and walk away.

3. The night shift

Every morning it takes the oldest approved idea and opens a pull request. One guide a day maximum, however long the queue gets, because thirty approved ideas turning into thirty posts in a week is how sites get flagged.

A real PR my agent opened overnight. I reviewed it, merged it, live the same day.

It builds free tools too. Small utilities related to your product: a calculator, a checker, a generator. Nobody links to your blog post about keyword research. People link to a free tool that does keyword research. And "free X checker" searches convert, because the searcher has a job to finish right now.

Guides bring the traffic. Tools convert it, and they earn links while you sleep.

A free tool my agent built and shipped. People bookmark tools. Nobody bookmarks your 2,000-word guide.

The agent never touches the live site. It proposes, you merge. That's the whole safety model, and it's also why the output doesn't rot into slop. You're still the editor.

4. Three gates that stop it writing slop

The part I'd want to read the source of before trusting it:

The SERP gate. It reads page 1 before writing a word. Whatever those results cover becomes the must-cover list. If the draft can't beat page 1, it refuses to build. An SEO tool that refuses to write.
Sameness detection. Every guide gets checked against your own back catalogue: opening patterns, heading skeletons, stock phrases. Write 30 articles and they converge on one template, and that's the scaled-content fingerprint Google discounts. Surfer and Clearscope compare you to competitors, never to yourself.
Information gain. Before it writes, it needs one thing no competitor page has. A command it actually ran, a real measurement, a screenshot from your dashboard.

Feed it what only you know: product knowledge, screenshots of your app, the words your customers actually use. That's the entire difference between content and slop.

What it costs

Search Console only — $0. Rankings from GSC, keyword ideas from Autocomplete and your own impression data.
Plus a free SerpApi key — still $0. Live page-1 checks, 250 free searches a month.
Plus DataForSEO (optional, most accurate) — cents per call. Search volume, keyword difficulty, domain rating.

Free mode finds keywords you can win. Paid mode also tells you which ones are worth winning. The agent itself runs on the Claude subscription you already pay for, and one PR a night is a small slice of it.

Does it work? Here are my charts

My site is very young. Two months old, no domain rating, no bought links: 84 clicks and 4.3K impressions in the last 60 days.

My site's Search Console, last 60 days. 84 clicks, 4.3K impressions, and climbing.

Anyone who's done SEO knows how to read that. A two-month-old domain with no DR is usually still sitting at zero, and the standard advice is to expect nothing for the first six months. This one didn't wait.

Why should you listen to me?

I build DispatchSEO, and all of it runs in public. The repo is open source, the sites my agent manages are live, and the search data above is mine. Last week I posted this same chart one week younger. Next month I'll post it again, whichever way it went.

You don't have to trust me. You can read the commits.

github.com/NeoZi12/dispatchseo

Everything in this article runs from this repo. Read the commits.

If you'd rather not run a machine, there's a hosted version at dispatchseo.com. Self-hosting stays feature complete either way. The cloud sells convenience, not capability.

What job should your agent take over next? I'm collecting ideas for the next article. Tell me below 👇

how to make money with GPT-5.6 Sol so fast it feels like cheating 🤖💰

Neo Zino — Wed, 05 Aug 2026 09:04:35 +0000

Two weeks after launch, OpenAI confirmed something wild: during an internal security eval, GPT-5.6 Sol escaped its sandbox and got into Hugging Face's production infrastructure looking for the benchmark answers. OpenAI called it an "unprecedented cyber incident."

That's the model you can now rent for $5 per million tokens.

OpenAI's own disclosure. The model hacked a third party during its own safety eval.

OpenAI shipped three models on July 9:

GPT-5.6 Sol ($5 in / $30 out per million tokens). The flagship. 1M context, built for long agentic runs.
GPT-5.6 Terra ($2.50 / $15). The workhorse.
GPT-5.6 Luna ($1 / $6). The cheap one. OpenAI claims it beats Claude Opus 4.8 at a quarter of the cost.

The whole GPT-5.6 family. The flagship costs $5 per million tokens in.

Building is now the easy part, distribution is now the bottleneck.

Here is how to use next-gen AI to solve the new bottleneck:

1. Build your SaaS

For years the advice was "don't build in a crowded market, you'll get crushed." That advice assumed building takes months. It doesn't anymore.

Sol is an execution beast:

Nick Saraev gave it image gen, video gen and hosting through MCPs. It created 80+ products with stills and video ads and published a showcase site. In about 10 minutes.
Nate Herk got a finished, edited video in his own cloned voice from one prompt. No human touched a timeline.

One prompt in, a finished edited video out.

So build the thing. Pick the crowded market you know best, ship the version with an agentic angle (a CLI your users' agents can drive, an MCP server, an API-first core), and get to distribution while your competitors are still writing their spec.

One warning before you burn money: Ultra mode spawns sub-agents that inherit Ultra, and your budget evaporates. Herk's video burned 458M tokens, about $318. Launch week was full of "hit my $200 limit in 4 hours" posts. Ultra is a demo setting. Run High.

2. Growth hack your SEO

The game is distribution, and SEO is the channel where agents have the biggest unfair advantage right now.

Most people paste "write me a blog post" into ChatGPT and publish it. It doesn't rank, because Google has read a billion pages exactly like it.

The growth hack is running the whole SEO operation as an agent loop, with the models routed by cost:

Luna does the grunt work: keyword lists, clustering, competitor checks. Thousands of calls, costs almost nothing.
Terra writes the drafts.
Sol only gets the judgment calls: what to build next, what's missing from the top 10.

Step by step:

1. Install Codex CLI and pick your model:

npm i -g @openai/codex
codex
# then inside: /model → gpt-5.6-sol, effort High

2. Turn your Agent into your SEO Manager.

Register for DispatchSEO, it gives your agent tools like: keyword tracking, Search Console data, a queue the agent fills, and every article arrives as a pull request. Works with Codex and Claude Code. It's also open source, so you can self-host it for free.

My agent's daily report. One guide a day, and it's ahead of schedule.

3. Run the setup wizard: connect Google Search Console (free) and your site's GitHub repo. Now the agent reads real queries and impressions, not guesses.

4. Turn on the schedule: weekly keyword research on Luna money, daily builds where Sol ships ONE thing as a PR. You review, you merge.

A real PR my agent opened. I reviewed, merged, and the tool went live the same day.

5. Feed it what only you know: product knowledge, screenshots of your app, the exact words your customers use. That's the entire difference between content and slop.

6. Scale more. Connect your other sites, one DispatchSEO install manages all of them, and the same agent works every queue.

I run this loop on my own sites. A 2-month-old, no-DR site pulled 60+ Google clicks and 3.2K impressions in the last 30 days. My newer site is 19 days in: 6 guides and 3 free tools shipped last week, and impressions nearly tripled in seven days, 611 to 1,785.

ClockedCode's Search Console, last 30 days. 61 clicks, 3.19K impressions, and climbing.

3. Build free tools that earn links while you sleep

Small free tools are backlink magnets. Nobody links to your blog post about keyword research. People link to a free tool that does it. And "free X checker" searches convert, because the searcher has a job to finish right now.

A free tool for Claude Code users, built by my Codex agent. Traffic is traffic.

The recipe:

Use Luna to mine tool-shaped keywords: "X generator", "X checker", "X calculator". Low difficulty, clear intent.
Have Sol build it as a single page. Free, no signup wall. The tool IS the marketing.
Put it on your main domain, so the links feed the site you're trying to rank.
Link to it from every related article, and from it back to your product exactly once.

One good free tool out-earns ten blog posts, and it now costs a prompt and a code review.

4. Sell the result, not the tool

Remember play 1: building got easy. That cuts both ways. If anyone can build the tool, the tool stops being what people pay for. People pay for the result.

Danny Postma's HeadshotPro is the perfect example. Underneath it's an image model anyone can run, and there are a hundred free "AI photo" tools. He doesn't sell any of that. He sells "professional headshots by tomorrow, no photographer, $29", and that framing built a seven-figure business while the model underneath became a commodity.

Not one word about the model underneath. Just the result and 3,518 reviews.

Why should you listen to me?

I build DispatchSEO, the tool from play 2, and everything in this article runs in public. The repo is open source, the sites my agent manages are live, and the search data above is mine. In a month I'll post the next chart, whichever direction it goes.

You don't have to trust me. You can read the commits.

Everything in this article runs from this repo. Read the commits.

Run these plays in order and they stack: the visitors compound month after month, and if you sell the result instead of the tool, those visitors turn into revenue.

Sol made building cheap. Distribution is where the money moved. Point your agent there.

If any of this sounds useful, try it yourself, the open-source version costs nothing to self-host.

Or take the hosted route if you don't want to touch Docker, and see what your chart does in a month.

How I Turned my AI agent Into an SEO Manager Using MCP (open-source)

Neo Zino — Mon, 03 Aug 2026 11:48:24 +0000

Every day i would do the same SEO routine: find keywords, check google results, write an article, publish it, track rankings.

So i thought why not turn my agent, the same agent that knows my product inside out, into my own SEO Manager!

And it works... so why not open-source it?

I made a dashboard to control and track everything, an MCP server to connect the dashboard to my agent

How it's built:

The agent (Claude Code or Codex, your pick) runs inside GitHub Actions on a cron. (it's self-hosted so BYO subscription)
The backend is an MCP server, there's no AI in it at all, it's just a database with a door on it.
Most of the MCP tools are just reads and writes to Postgres. A few pass through to DataForSEO for keyword and SERP data.
The agent does the thinking, the server does the remembering.
The agent fetches its instructions from the server when the run starts.
It never writes to the site directly. Always a PR, with checks run against it first.
Works with Claude Code and Codex coding agents (more to come)

Key features:

research keywords
analyze rankings
create content workflows
monitor search performance

Fully open-source (AGPL-3.0)

Open-source repo link: https://github.com/NeoZi12/dispatchseo

Would love feedback from people building with agents

I paid for Ahrefs and did SEO by hand every day. So I gave the job to Claude Code.

Neo Zino — Tue, 28 Jul 2026 15:54:05 +0000

For a long time I ran SEO for my own product the normal way. Ahrefs subscription, plus an hour or two most days writing guides, plus a chunk of every week building small tools to rank for long-tail keywords. It worked. It also ate a huge slice of my day, every single day.

At some point I was staring at another keyword list, about to write another guide I'd already half-written in my head a dozen times, and it hit me: I already use Claude Code to build this product. Why am I the one doing this part by hand?

Claude Code already knows my product better than I do, it wrote a lot of the codebase. A tool like SEObot or Outrank has to crawl your homepage and guess what you do. My own coding agent doesn't have to guess. Give it the right tools, real keyword data, rank tracking, a way to actually ship content, and it should do this job better than I was doing it manually.

That became DispatchSEO. It's an MCP server plus a dashboard. The MCP server is what Claude Code connects to: keyword research, rank history, Search Console stats, backlink prospects, a queue of content ideas. The agent reads all of that, decides what to research or write next, and opens a pull request when it's got something real. The dashboard is where I sit and approve or reject what it queues up. I stopped writing the guides. I didn't stop deciding what ships.

It's 100% free and open source (AGPL), and self-hosting is one command if you've got Docker

Everything the paid cloud version does, the self-hosted version already does, in your own Vercel/Supabase/GitHub accounts, for $0. You bring a free Google Search Console connection; a DataForSEO account is optional if you want real search volumes instead of the free GSC-only mode.

I built this because I was the exact user it's for: running a solo product with no time to also run an SEO team. If that's you too, the self-host path costs a Docker install. I'd honestly like to find out what breaks the first time someone else's Claude Code runs it instead of mine.  Highly recommended to use a $5 VPS for this

Repo (100% free, open source, AGPL): github.com/NeoZi12/dispatchseo

How to Use Claude Code in VS Code: The Complete Setup I Use Every Day

Neo Zino — Thu, 09 Jul 2026 09:59:16 +0000

Originally published on ClockedCode - my blog about getting the most out of Claude Code.

There are two ways to use Claude Code in VS Code: the official extension, which gives you a graphical panel inside the editor, and the CLI running in the integrated terminal. I use it every single day to build my products, and the full setup takes about ten minutes. This guide walks through every step, then the setup problems people actually hit - I hit most of them myself.

TL;DR: Install the free Claude Code extension from the VS Code Marketplace (publisher: Anthropic), sign in with a Claude Pro/Max plan or Console credits, then add the CLI so claude also runs in the integrated terminal. Run /init to generate a CLAUDE.md, /terminal-setup to fix Shift+Enter, and /statusline to see your context at a glance. That is the whole setup I ship real products with.

Extension vs terminal: what "Claude Code in VS Code" means

This trips up almost everyone, so first things first. "Claude Code VS Code" can mean two different things, and both are official:

The Claude Code VS Code extension. A native extension by Anthropic (listed as "Claude Code for VS Code" on the Marketplace) that adds a chat panel to your editor - a spark icon in the top bar, inline diff reviews, quick fixes from the editor's lightbulb menu, plan documents that open as tabs, file mentions with autocomplete, and drag-and-drop images in chat. If you searched for the extension, this is it.
The Claude Code CLI in the integrated terminal. The same agent, running as a command-line program inside VS Code's built-in terminal. You get the full command set and every flag, and when the extension is installed, the CLI connects to the editor so diffs open in VS Code's native diff viewer.

They are not competitors. The extension bundles its own copy of the CLI for the chat panel, and both read the same account, the same settings files, and the same CLAUDE.md. My honest recommendation after months of daily use: set up both. I live in the extension day to day, and keep the CLI installed for the handful of commands that only exist there.

Prerequisites

Three things, and only the second one costs money:

VS Code 1.98.0 or later. Check with Help, then About. Anything from 2025 onward is fine.
A Claude account on a paid plan. Claude Pro or Max (the subscription most individual devs use), Team or Enterprise, or a Claude Console account with pre-paid API credits. The free Claude.ai plan does not include Claude Code access - there is no free tier for this tool.
Node.js 22+, but only for the npm install route. The extension bundles everything it needs, and the native CLI installers ship a standalone binary. You only need Node if you specifically want to install the CLI with npm.

That's it. No API key juggling - signing in through your browser handles authentication for both the extension and the CLI.

Installing the Claude Code VS Code extension

The extension is the fastest way in, so start here:

Open the Extensions view: Cmd+Shift+X on Mac, Ctrl+Shift+X on Windows/Linux.
Search for Claude Code and pick the official Claude Code for VS Code extension published by Anthropic (verified badge). The results are full of similar-looking third-party extensions - "Chat for Claude Code", "Claude Code Assistant", and friends. Those are not Anthropic's; check the publisher before you click.
Click Install.
Open any file in your project (the panel needs a file open, not just a folder), then click the spark icon in the top bar - or press Cmd+Esc (Ctrl+Esc on Windows/Linux).
Sign in when prompted. It opens your browser for a one-time OAuth login with your Claude account, then remembers you.
Type a first prompt like "explain what this file does" to confirm everything is wired up.

Using Cursor or another VS Code fork? The extension installs there too - search the fork's Extensions view, or grab it from the Open VSX registry. And if a fork ever misbehaves, the CLI in the integrated terminal works everywhere.

Setting up Claude Code in the VS Code terminal

Here is the part most guides skip: the extension does not put the claude command on your PATH. If you want to run Claude Code in the integrated terminal - and you eventually will, because that is where the full command set lives - you need the standalone CLI too.

Install the CLI with the native installer for your OS (no Node.js required):

# macOS / Linux / WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows (PowerShell)
irm https://claude.ai/install.ps1 | iex

Prefer a package manager? brew install --cask claude-code on macOS, winget install Anthropic.ClaudeCode on Windows, or npm install -g @anthropic-ai/claude-code if you have Node 22+.

Open the integrated terminal (Ctrl+`) and run:

claude

If it asks you to sign in, run /login - same browser flow as the extension, and if you already signed in through the extension you are usually done.
Because you launched it inside VS Code with the extension installed, the CLI connects to the editor automatically: diffs open in VS Code's native viewer, and your current selection is visible to Claude as context. Launched from an external terminal instead? Run /ide inside the session to connect manually. (Git Bash users: see the first item in troubleshooting.)
Run /terminal-setup once. It configures keybindings so Shift+Enter inserts a newline instead of sending the message - the single most annoying default in terminal land, fixed in two seconds.

Essential configuration

Do this before your first real task. A stock install works; a configured install is a different tool entirely, and it costs ten minutes.

Generate a CLAUDE.md with /init. If you are new to Claude Code, this is the one concept worth actually learning. CLAUDE.md is a plain markdown file at the root of your project where you write, in normal English, everything you would tell a new developer on their first day: how to run the app, how the project is structured, which conventions to follow, what never to touch. Claude Code reads it automatically at the start of every session - so instead of re-explaining your project in every conversation, you explain it once and Claude just knows. Run /init and it generates a starter version by reading your codebase; then trim it, because short and specific beats long and vague, and this file rides along in every conversation. I wrote a full breakdown of what belongs in a CLAUDE.md, with my real file as the example.

Know your three settings files. User-wide settings live in ~/.claude/settings.json, per-project settings in .claude/settings.json (committed to git), and personal per-project overrides in .claude/settings.local.json (gitignored). Add "$schema": "https://json.schemastore.org/claude-code-settings.json" at the top and VS Code gives you autocomplete for every option. For everything else there is /config, which opens a settings interface right in the session.

Set up the status line with /statusline. One command, and Claude Code permanently shows your model, git branch, and how much of the context window you have used. Context awareness is the difference between a sharp session and a degraded one, so make it visible.

Add MCP servers - but only the ones you will use daily. MCP servers connect Claude Code to tools like GitHub, Supabase, and Sentry. They are genuinely great and also the easiest way to silently burn your context window, since every connected server costs tokens on every turn. I keep a curated list of the six MCP servers actually worth installing - resist the directory-of-fifty approach.

One personal tip before you move on. The single best line in my global CLAUDE.md is a rule that Claude must end every response with a one-line TL;DR summary. Long outputs stopped being homework: I read the one line, decide whether the details matter, and move on. Steal it - add "End every response with a TL;DR line, one or two sentences max, summarizing what was done and what I need to decide next" to your CLAUDE.md and feel the difference the same day.

Daily workflow: how I actually use it

The install is the easy part. Here is what makes Claude Code in VS Code productive day to day:

Plan before code, always. For anything bigger than a one-line fix, I have Claude propose a plan first and only then let it touch files. It is the single highest-leverage habit with this tool - I wrote up exactly how I ask Claude Code to plan before it codes.
Run parallel sessions instead of waiting. This is the biggest unlock in my day. Claude thinking is not a coffee break: I keep a few Claude Code sessions open side by side in VS Code (Cmd+Shift+Esc opens a new one in its own tab) and rename each one to its job - one builds the feature, one chases a bug, one answers questions about the codebase. While one session works, I jump to another. I stopped waiting for Claude months ago; you should too.
Give parallel sessions their own git worktree. Two sessions editing the same files will happily overwrite each other's work. Git worktrees fix that: git worktree add ../myapp-feature -b feature-x gives a session its own folder on its own branch, backed by the same repo. Open that folder in a second VS Code window, point a Claude Code session at it, and the parallel work can never collide - merge the branches when they are done.
Pick your permission mode - I run Auto. Out of the box, Claude Code asks before it touches a file or runs a command (manual mode). Flip the toggle in the panel to Auto mode and it works straight through without stopping to ask every time. I run Auto all day because I trust Claude at this point - but that trust stays cheap only because of the next habit.
Review every diff like a pull request. Changes open in VS Code's diff viewer. Read them. The tool is excellent; it is not infallible, and the diff view exists precisely so you stay the reviewer. Keeping the work on a feature branch makes that oversight cheap.
Keep an eye on your usage. /usage shows where your quota goes (including a breakdown by MCP servers and subagents), and /context shows what is eating the window. If you are hitting limits faster than you should, scan your setup with UsageCut - my free tool that flags what is silently burning tokens - and read why Claude Code eats usage limits so fast.

And my real split, for what it is worth: I am in the extension 99% of the time. The terminal gets the remaining 1%, purely for the commands that only exist in the CLI. If you were expecting a sophisticated two-tool workflow, sorry - the panel is simply where the work happens, and the CLI sits there for the rare command the panel does not surface.

Troubleshooting

Every one of these is a real, documented failure mode, in order of how often they bite:

"No available IDEs detected" when connecting from the integrated terminal. The most common integration failure, especially on Windows with Git Bash: non-default shells hide the environment variables the CLI uses to detect the running VS Code session. The community workaround is to clear them when launching - TERM_PROGRAM= claude, or the fuller VSCODE_PID= VSCODE_CWD= TERM_PROGRAM= command claude (alias it if you hit this daily). See the GitHub issue where this workaround comes from.
The spark icon is missing. You need a file open, not just a folder - the icon does not render in an empty workspace view. Still missing? Check VS Code is 1.98.0+, reload the window (Cmd+Shift+P, then "Developer: Reload Window"), and look for the Claude Code item in the status bar (bottom right) as an alternative entry point.
Cmd+Esc does nothing on macOS. On newer macOS versions (Tahoe and later), the system Game Overlay intercepts that exact shortcut. Fix: System Settings, Keyboard, Keyboard Shortcuts, Game Controllers - uncheck the Game Overlay binding. Or rebind Claude Code's shortcut in VS Code's Keyboard Shortcuts.
claude: command not found after installing the CLI. The native installer puts the binary in ~/.local/bin, which may not be on your PATH. Add export PATH="$HOME/.local/bin:$PATH" to your ~/.zshrc or ~/.bashrc and restart the terminal.
The extension refuses to work in a new folder. Check workspace trust: Claude Code does not run in VS Code's Restricted Mode. Click the shield icon in the status bar and trust the workspace.
Windows: native, WSL, and Git Bash all behave differently. Native PowerShell and CMD work, and WSL works - but if your projects live in WSL, install and run the CLI inside WSL (use the Linux install command), and with VS Code Remote - WSL, install the extension into the WSL environment, not just the Windows side. Git Bash is the quirky one: it usually needs the environment-variable workaround from the first item above.
Login loops or auth errors. Run /logout, then /login again. On WSL or SSH, the browser may hand you a code instead of redirecting automatically - paste it back into the terminal.
Shift+Enter sends the message instead of adding a line. That is the un-fixed terminal keybinding - run /terminal-setup once and it is gone.

FAQ

Does Claude Code work in VS Code?

Yes. There are two official ways: the free Claude Code extension by Anthropic (a graphical panel inside VS Code) and the Claude Code CLI running in the integrated terminal. Both are fully supported, share the same account and settings, and work in VS Code 1.98.0 or later.

Is the Claude Code VS Code extension free?

The extension itself is a free download from the VS Code Marketplace, but using it requires a paid Claude plan: Pro, Max, Team, Enterprise, or a Claude Console account with API credits. The free Claude.ai plan does not include Claude Code access.

Claude Code extension vs terminal: which is better?

Start with the extension: it is easier to learn and shows diffs, plans, and file mentions in a graphical panel. Keep the CLI installed too, for the full command set and the flags the panel does not surface. I use the extension about 99% of the time, and both share the same configuration.

Do I need Node.js to use Claude Code in VS Code?

Not anymore. The extension bundles everything it needs, and the native CLI installers for macOS, Linux, and Windows download a standalone binary. Node.js 22 or later is only required if you install the CLI through npm.

Does Claude Code work in Cursor and other VS Code forks?

Yes. The extension installs in Cursor and other VS Code forks - search the fork's Extensions view or install it from the Open VSX registry. The CLI works in any editor's integrated terminal regardless of fork.

Does Claude Code do inline autocomplete like GitHub Copilot?

No. Claude Code is an agent, not an autocomplete engine: instead of completing code token by token as you type, it takes a task, works across files, and proposes diffs you review in the editor. The two tools solve different problems, and plenty of developers run both.

How do I open Claude Code in VS Code?

Press Cmd+Esc on Mac or Ctrl+Esc on Windows and Linux to toggle focus between the editor and Claude Code, or click the spark icon with a file open. Cmd+Shift+Esc (Ctrl+Shift+Esc on Windows and Linux) opens a new conversation in its own tab.

You're set up - now make it yours

That is the complete setup: extension installed, CLI in the terminal, CLAUDE.md generated, keybindings fixed, status line on, and a workflow that treats Claude Code as a reviewed collaborator instead of a magic button. The gap between developers who get real leverage from Claude Code in VS Code and those who bounce off it is almost never the model - it is this configuration and the habits on top of it. Take the ten minutes.

If this helped, the original post stays updated as Claude Code changes - and I share free Claude Code tools and the full setup kit at ClockedCode.