기본 콘텐츠로 건너뛰기

Cline Tutorial: Plan, Act and Roll Back Safely in VS Code

Summary: Cline is an open-source AI coding agent that runs inside VS Code (and Cursor, Windsurf, JetBrains and the terminal). It can read your project, edit files and run commands, but only after you approve each step. This how-to walks through installing it, choosing how to pay for the model behind it, and using the three features that make it safe to hand real work to: Plan and Act modes, checkpoints, and carefully limited auto-approve.

What Cline is, and what it isn't

Cline is an agent rather than an autocomplete tool. You describe a task in a chat panel, and Cline decides which files to read, proposes edits as diffs, and asks to run terminal commands. According to its documentation, every file creation, file edit and terminal command needs your explicit approval unless you change that yourself. The extension's source code is published on GitHub under the Apache 2.0 licence, so the tool itself costs nothing.

What you do pay for (or not) is the AI model that does the thinking. Cline doesn't ship with one model. You connect a "provider", which can be Cline's own pay-as-you-go service, its ClinePass subscription, your own API key from a company like Anthropic, OpenAI, Google, DeepSeek or OpenRouter, or a model running locally on your machine. That choice decides your costs far more than anything else, so it gets its own section below.

Cline is also not a sandbox. It works on the real files in the folder you open, and the commands it runs are real commands on your computer. The habits in this guide exist because of that.

Step 1: Install Cline in VS Code

  1. Open the Extensions view with Ctrl+Shift+X (Windows/Linux) or Cmd+Shift+X (Mac).
  2. Search for Cline and click Install. The official listing on the VS Code Marketplace, linked from Cline's docs, has the ID saoudrizwan.claude-dev, so check for that ID rather than installing a lookalike.
  3. Click the Cline icon (a small robot) in the Activity Bar to open the panel. If you can't find it, open the Command Palette and run Cline: Open In New Tab.

The same steps work in Cursor, Windsurf, VSCodium and Antigravity. Windsurf and VSCodium pull the extension from Open VSX, but the flow is identical. For JetBrains IDEs, install Cline from Settings → Plugins → Marketplace, restart, and open it from View → Tool Windows → Cline. If you live in the terminal, the CLI needs Node.js 20 or later (22 is recommended) and installs with npm install -g cline, followed by cline auth.

One optional tweak from the official tutorial is worth doing on day one: by default the Cline panel replaces your file explorer on the left. Drag the Cline icon into VS Code's secondary sidebar on the right (toggle it with Option/Alt + Cmd/Ctrl + B) so you can watch files change while you chat.

Step 2: Choose how you'll pay for the model

Open Cline's settings with the gear icon, pick an option in the API Provider dropdown, sign in or paste a key, and choose a model from the Model dropdown. Here is how the three main routes compare, using only what Cline's documentation states as of today:

Option Price Free use and limits Key features
Cline (usage-billing) Pay as you go with Cline credits, added from the Cline dashboard Models tagged FREE in the selector are part of rotating, limited-time promotions with a usage quota; they work only in the IDE extension and CLI, not through the Cline API, and that usage may be used to improve models One sign-in with Google, GitHub or email; no API keys to manage; 100+ models on one balance; usage shown under Settings → View Usage
ClinePass $9.99 per month flat Usage is measured against a rolling 5-hour window, a weekly limit and a monthly limit; free promo models can also appear here during promotions A curated set of open coding models (GLM, Kimi, DeepSeek, MiMo, MiniMax, Qwen and others) with 2–5x the usage you'd get at standard API rates; optional and usable alongside other providers
Bring your own key / local Whatever your provider charges; local models have no per-request cost Depends on your provider's own free tier, if any; local models depend on your hardware (Cline suggests 16–32 GB RAM for small quantized models, 64 GB+ for larger ones) Use an existing Anthropic, OpenAI, Gemini, DeepSeek, Bedrock or OpenRouter account, a Claude Pro/Max subscription via Claude Code, or Ollama / LM Studio running on your machine

A sensible starting order for a newcomer: try a FREE-tagged model under the Cline provider to learn the workflow, then decide. If you already pay for an API, BYOK keeps everything on one bill. If you want predictable spending on open models, ClinePass is the flat-fee option. If privacy matters most, run a local model, and in that case Cline's docs recommend turning on Use Compact Prompt under Settings → Features and keeping tasks small.

For local setups, the Ollama provider expects http://localhost:11434 and LM Studio expects http://localhost:1234, with LM Studio's server started from its Developer tab. Our LM Studio tutorial covers getting a model loaded first. If you'd rather use one API key for many cloud models, our OpenRouter free models breakdown explains its daily caps, which apply when Cline talks to OpenRouter too.

Step 3: Open a folder and plan before you build

Always open a project folder first (File → Open Folder). The docs warn that without one, Cline doesn't know where to create files. For practice, an empty folder is ideal.

At the bottom of the Cline panel there's a toggle that reads Plan or Act. The difference matters:

  • Plan mode lets Cline read your code, run searches and discuss an approach, but it cannot change files or run commands.
  • Act mode lets Cline edit files and run commands, still asking for approval, and it keeps the whole conversation from Plan mode, so you don't have to repeat yourself.

The workflow Cline's docs recommend scales with task size:

  1. Small, obvious fixes (a typo, a missing import, a config value): go straight to Act.
  2. Medium tasks (a new endpoint, a UI component, a bug that needs investigating): describe the goal in Plan mode, point Cline at the relevant files with @/path/to/file mentions (a trailing slash, as in @/src/components/, pulls in a folder), agree on the approach, then flip to Act.
  3. Large, multi-file work: type /deep-planning. Cline investigates the codebase, asks clarifying questions, writes an implementation_plan.md, and then creates a task to carry it out.

A cost tip that is easy to miss: in settings you can enable Use different models for Plan and Act, pairing a stronger reasoning model for planning with a cheaper or faster one for the edits. Cline switches models automatically when you switch modes.

A first prompt to try, adapted from the official tutorial: ask Cline, in Act mode, to build a todo app in a single HTML file with an input field, a list, checkboxes with strikethrough, delete buttons and inline CSS and JavaScript. You'll see the proposed file as a diff with green additions and an Approve / Reject pair underneath. Approve it, open index.html in a browser, then ask for a follow-up change in the same chat, such as saving tasks to local storage, and watch Cline edit the existing file instead of starting over.

Step 4: Use checkpoints as your undo button

Checkpoints are on by default. After every tool use, Cline commits a snapshot of your files to a separate "shadow" Git repository, so your own Git history stays untouched and even files Git doesn't track are captured. In the chat you'll see a Checkpoint marker with two buttons:

  • Compare opens your editor's diff view for that step.
  • Restore offers three choices: Restore Files (undo the code, keep the conversation), Restore Task Only (keep the code, delete later messages so you can re-prompt), and Restore Files & Task (go back completely).

The practical rule: if the conversation was useful but the code broke, restore files and ask for a different approach. If the code is fine but the chat went off course, restore the task only. Checkpoints are a safety net, not a replacement for your own commits, so commit whenever you reach a version you like.

Step 5: Teach it your rules and manage long tasks

If you keep typing the same instructions ("use TypeScript", "don't touch /legacy"), put them in a rules file. Click the scale icon at the bottom of the Cline panel, choose New rule file..., and write plain markdown. Workspace rules live in .clinerules/ (or .cline/rules/) and can be committed for your team. Cline also detects existing .cursorrules, .windsurfrules and AGENTS.md files, and each rule can be toggled on or off. The /newrule command can draft one for you.

Long sessions fill the model's context window, which makes responses slower and costlier. Two commands help: /smol (alias /compact) condenses the current conversation so you can continue, and /newtask starts a fresh task carrying over only the plan, progress, relevant files and next steps.

Step 6: Loosen auto-approve gradually

Clicking Approve on every file read gets tedious, and Auto Approve lets you skip prompts per category: reading project files, editing project files, running "safe" commands, using the browser and using MCP servers. Two details from the docs are worth knowing. First, the "all files" variants extend access outside your workspace and only work if the matching base toggle is on. Second, there is no fixed allowlist of safe commands: the model itself flags each command as needing approval or not, so "safe" is a judgement, not a guarantee.

Cline's own recommended default is to auto-approve Read project files only, and to leave edits, commands, browser and MCP off until you have a reason. If you do allow edits, keep checkpoints on. YOLO mode, under Settings → Features, approves everything, including destructive commands and Git pushes, with no confirmation. Keep it for throwaway folders.

Editor's note

Who it's for: developers and technical hobbyists who want an agent that can work across many files but still asks before acting, and anyone who wants to choose (or switch) the model and the bill behind their coding assistant. Not ideal for: people who only want inline autocomplete while typing, or anyone unwilling to review diffs and commands before approving them.

Gotchas I verified in the docs: FREE-tagged models are limited-time promotions with a quota, not a permanent free plan, and their usage may be used to improve models. .clineignore is being deprecated because it never stopped Cline reading files through shell commands, so it was never a security boundary; keep secrets out of the workspace instead. Checkpoints can use a lot of storage and slow Cline down in very large repositories; the docs suggest turning them off if you notice that. Using a local model with the default prompt may struggle, which is why the docs recommend Compact Prompt.

Common errors and fixes: "Unauthorized: Please sign in" means your session expired, so click Sign In again. The sign-in browser doesn't open: copy the URL from the Cline output panel by hand. Local model won't connect: make sure Ollama or the LM Studio server is running, check the base URL, and confirm the model is pulled or loaded. Can't reach providers at work: the VS Code extension uses VS Code's own proxy settings, while the CLI reads the standard https_proxy/http_proxy variables (and NODE_EXTRA_CA_CERTS for a custom certificate).

FAQ

Is Cline free?

The extension is free and open source. Model usage is what costs money: FREE-tagged promotional models cost nothing until their quota runs out, local models cost nothing per request, and everything else is billed by Cline credits, ClinePass ($9.99/month) or your own provider.

Can I use Cline with my Claude or ChatGPT subscription?

Cline's provider list includes Claude Code for Claude Pro/Max subscribers and an OpenAI option that supports either an API key or OpenAI Codex sign-in. Check the provider page for your plan before relying on it, since what each subscription allows is set by that company.

How is Cline different from GitHub Copilot's free tier?

Copilot is tied to a GitHub account and GitHub's plans, while Cline lets you pick any supported provider and puts an approval step in front of every action. If you want to compare, our GitHub Copilot Free in VS Code guide covers that setup.

Start in Plan mode on a small, committed project, keep checkpoints on, and only widen auto-approve once you trust what you're seeing.

See the official screenshots at the end of this post. Image credit: Cline Docs (docs.cline.bot) and our own screenshot of the public Cline checkpoints documentation page.

Last verified: 2026-10-27

Sources:
Cline Docs: Overview
Cline Docs: Installing Cline
Cline Docs: Authorization
Cline Docs: Cline (usage-billing)
Cline Docs: ClinePass
Cline Docs: Free Models
Cline Docs: IDE tutorial
Cline Docs: Plan & Act Mode
Cline Docs: Checkpoints
Cline Docs: Auto Approve & YOLO Mode
Cline Docs: Rules
Cline Docs: Using Commands
Cline Docs: Local models
Cline Docs: Networking and Proxies
Cline Docs: Deprecations
Cline on GitHub

Uploaded Image Uploaded Image Uploaded Image Uploaded Image

댓글

이 블로그의 인기 게시물

브루(Vrew) 요금제 정리: 무료 200크레딧으로 어디까지 쓸 수 있나

요약: 영상 자막 편집기 브루(Vrew)는 2026년 4월 22일부터 기능별 사용량 대신 '통합 크레딧' 하나로 요금을 매겨요. 무료 플랜은 매달 200크레딧이 들어오는데, 음성 분석(자동 자막)만 쓰면 최대 200분 분량이라 짧은 영상 몇 편은 충분하지만, 영상 초반의 Vrew 마크를 지우려면 유료 플랜이 필요해요. 이 글은 브루 공식 가격 페이지와 고객센터 공지를 2026년 10월 23일에 다시 확인해, 크레딧 계산법과 플랜 고르는 기준을 정리했어요. 목차 브루 요금 구조: 통합 크레딧이란 1크레딧으로 할 수 있는 일 무료 플랜으로 되는 것과 안 되는 것 유료 플랜 4종 비교표 월 구독·연 구독·기간 이용권, 무엇이 이득일까 상황별 추천: 학생·직장인·유튜버 크레딧이 모자랄 때: 사용 기록 지우개 업그레이드·환불 전에 알아둘 규칙 운영자 메모, 자주 묻는 질문 1. 브루 요금 구조: 통합 크레딧이란 예전 브루는 플랜마다 음성 분석 몇 분, AI 목소리 몇 자, 번역 몇 자처럼 기능별 한도를 따로 줬어요. 그런데 보이저엑스(브루 운영사)가 실제 이용 데이터를 보니 한두 가지 기능만 집중해서 쓰는 사람이 훨씬 많았다고 해요. 그래서 2026년 4월 22일 업데이트부터는 플랜마다 정해진 크레딧을 받고, 어떤 기능에 쓸지는 내가 정하는 방식으로 바뀌었어요. 플랜별 매달 지급량은 Free 200 · Light 2,000 · Standard 10,000 · Business 12,000크레딧 이에요. 공식 공지에 따르면 무료 사용자는 매달 1일에, 유료 사용자는 매월 결제일에 충전되고, 남은 크레딧은 다음 달로 넘어가지 않아요. 바뀐 뒤 무엇이 좋아졌나 공지에 나온 예시가 이해하기 쉬워요. Standard 플랜에서 AI 목소리만 쓰면 예전에는 최대 50만 자였지만, 지금은 1만 크레딧을 모두 AI 목소리에 몰아 쓸 수 있어 최대 100만 자가 돼요. 반대로 여러 기능을 골고루 쓰는 사람은 크레딧이 기능끼리 ...

에이닷 전화 vs 익시오: 통화 녹음·AI 요약, 내 통신사·폰에 맞는 앱 고르기

요약: 통화 녹음과 AI 요약을 쓰고 싶다면 먼저 내 통신사와 휴대폰 종류부터 확인해야 해요. LG유플러스 고객은 무료 앱 익시오, SKT 고객은 에이닷 전화가 가장 기능이 많고, 다른 통신사의 아이폰 사용자는 iOS에 들어 있는 기본 통화 녹음·전사 기능이 현실적인 선택이에요. 이 글은 2026-10-30에 SKT·LG유플러스·Apple 공식 페이지와 도움말을 직접 확인해 세 가지를 비교했어요. 왜 '앱 기능'보다 '내 조건'을 먼저 봐야 할까요 통화 요약 앱 비교 글을 보면 기능 목록부터 나열하는 경우가 많아요. 그런데 실제로 써 보려는 사람에게 가장 먼저 걸리는 건 기능이 아니라 이용 조건 이에요. 익시오는 LG유플러스 고객 전용이고, 에이닷 전화는 누구나 설치할 수 있지만 통역콜·baro 같은 기능은 SKT 고객에게만 열려 있어요. 아이폰은 운영체제 구조상 앱이 통화를 마음대로 녹음하기 어려워서, 통신사 앱이 안 되면 iOS 기본 기능에 기대야 해요. 그래서 이 글은 "어떤 앱이 제일 좋은가"보다 "내 통신사·내 폰에서 무엇을 쓸 수 있나" 순서로 정리했어요. 기능 차이는 그다음이에요. 한눈에 비교: 에이닷 전화 vs 익시오 vs 아이폰 기본 통화 녹음 구분 에이닷 전화 (SKT) 익시오 ixi-O (LG유플러스) 아이폰 기본 통화 녹음 (Apple) 누가 쓸 수 있나 통신사 구분 없이 앱 설치 가능. 단, 통역콜·baro·그룹통화 발신 등은 SKT 고객 전용 U+ 모바일 고객. 타 통신사, 만 14세 미만, 선불요금제 고객은 제한 (공식 사이트 FAQ상 알뜰폰 중 KB리브엠은 예외) 지원 국가·언어의 iPhone 사용자. 1명과 통화할 때만 녹음 가능 지원 OS·기기 Android 10 이상, iOS 16 이상 iOS 17 이상 iPhone 12 이상 / Android...

타입캐스트 vs 클로바더빙: 수익 나는 채널에 쓸 때 무료·유료 조건 비교

요약 타입캐스트와 클로바더빙은 둘 다 글만 입력하면 한국어 AI 목소리를 만들어 주는 서비스지만, 무료로 쓸 수 있는 범위가 꽤 다릅니다. 클로바더빙 무료는 매달 15,000자까지 쓸 수 있는 대신 수익이 생기는 채널에는 쓸 수 없고, 타입캐스트 무료는 처음에 주는 약 5분 분량의 다운로드로 끝나는 체험판에 가깝습니다. 광고 수익이 붙은 유튜브나 쇼핑몰 영상에 넣을 목소리라면 처음부터 유료 플랜 조건을 보고 고르는 편이 안전합니다. 왜 이 두 서비스를 같이 비교하나 쇼츠 내레이션, 수업 자료 더빙, 상품 소개 영상처럼 "내 목소리 대신 AI 목소리"가 필요할 때 한국 사용자들이 가장 많이 찾는 곳이 네오사피엔스의 타입캐스트 와 네이버의 클로바더빙 입니다. 기능 소개만 보면 비슷해 보이지만, 실제로 문제가 되는 건 음질보다 "이 목소리를 내 채널에 써도 되나" 입니다. 그래서 이 글은 기능 나열보다 무료 사용 범위, 출처 표기 의무, 유료 플랜 가격과 다운로드 한도를 중심으로 정리했습니다. 모든 수치는 2026년 10월 25일에 두 서비스의 공식 요금 페이지와 고객센터 도움말에서 직접 다시 확인한 내용입니다. 한눈에 보는 비교표 항목 타입캐스트 무료 타입캐스트 베이직 클로바더빙 Free 클로바더빙 Standard 월 요금 0원 9,900원 (연간 결제 시 10% 할인) 0원 19,900원 (네이버 클라우드 플랫폼에서 결제) 무료·포함 사용량 음성 생성·재생 무제한, 다운로드는 최초 3,000크레딧(약 5분) 1회 지급 월 30,000크레딧(약 60분) 월 15,000자, 다운로드 월 20회, 프로젝트 누적 5개 (매월 1일 갱신) 월 30,000자, 다운로드 월 40회, 프로젝트 누적 20개 쓸 수 있...