Claude Code Explained: Setup, Pricing, and Paying from Russia

In this guide we cover what Claude Code is and how it works, who it suits, how to install it and get started, what it costs, and what to do if paying for an Anthropic subscription from Russia is not an option. We also walk through a working alternative: running Claude Code without a subscription, on a Kumo balance topped up with a Russian card or via SBP.

Claude Code is not autocomplete and not a chat panel next to your editor — it is a full agent in the command line. You launch it in your project folder and it orients itself in the repository: scans the structure, opens the files it needs, finds where the logic you asked about actually lives. Then it does not just advise, it acts — editing files, executing shell commands, running builds and tests.

The key word is iteration. The agent makes a change, runs the tests, sees a failure, reads the output and fixes it, looping until the task is done or you stop it. Every change is shown to you for approval, so control stays with the human: you decide what lands in the codebase and what does not.

Because Claude Code lives in the terminal, it does not care what you write code in: VS Code, JetBrains, Vim, or no editor at all. That makes it universal — the same tool works on any stack and in any environment that has a console.

The core audience is developers who are comfortable in the terminal. If you already live in the console and run git and npm by hand, Claude Code slots into your workflow with zero friction: it is just one more command in a familiar environment, only a very smart one.

Beginners can use it too — with one caveat: the console should not scare you. The interface itself is a natural-language dialog; there are no flags or configs to memorize. If you can open a terminal, cd into a project folder and run a command, that is enough to start. If you would rather have a graphical editor with a built-in agent, Cursor or VS Code agents like Cline may fit better — we compared them in our overview of vibe-coding tools.

Installation is standard for a console tool: you need Node.js, then a single command — npm install -g @anthropic-ai/claude-code. After that the claude command appears in your terminal. Change into your project folder and run it — the agent starts right there and immediately sees your code.

On first launch Claude Code asks you to authenticate: either with an Anthropic account and subscription, or with an API key. We will get back to payment options below — for Russian-speaking users this is the painful part, and it has a solution.

The easiest way to start is reconnaissance: ask the agent to describe the project — the stack, the entry points, how things fit together. Then give it a small, concrete task: fix a bug, add form validation, write a test. Within half an hour you will understand how the agent thinks and where it needs more context from you. By default the agent does not perform potentially risky actions silently: it shows commands and edits and asks for confirmation, so breaking a project through inattention is hard.

The basic loop looks like this: you describe a task in words, the agent proposes a plan or goes straight to edits, you review the diff and accept it — or ask for a redo. The more specific the request, the fewer iterations: a task like add a change-password button to the profile page, styled like the other form buttons resolves much faster than just make it good.

Then the real iterative work begins. Ask the agent to run the tests after a change — it will read the output itself and fix whatever broke. A large feature is best cut into steps: data model first, then the API, then the UI, verifying after each step. It is the same process as code review, only much faster: you read diffs instead of typing the code yourself.

A tool of its own is the CLAUDE.md file in the project root. It holds the context the agent should always know: the stack and versions, code conventions, build and test commands, things that must never be done in this project. Claude Code picks this file up at the start of a session, so you do not repeat yourself in every conversation. Ten minutes spent on a good CLAUDE.md pay off on day one.

Claude Code's strength is the complete work cycle. Many tools can generate code, but here the agent also verifies its own work: it runs tests and commands, reads errors, and fixes them without your involvement. Combined with strong Claude models, this yields a rare quality: you can hand the agent an entire task, not just ask for code snippets.

That said, Claude Code is not the only player. In our overview of vibe-coding tools we covered the whole class: editors like Cursor, VS Code agents, terminal CLIs. And if openness matters to you, look at OpenCode — an open-source terminal agent with a similar philosophy; we have a separate article about it. The good news is you do not have to choose forever: these tools share compatible approaches, and the skill of working with an agent transfers.

Officially there are two ways to pay. The first is an Anthropic subscription: a fixed monthly fee, with usage limits depending on the plan. The second is API pay-per-token: you pay for the actual volume of text processed, which can be cheaper than a subscription under moderate load and more expensive under heavy load. There is no single price for Claude Code: it all depends on how much you work and on which models.

With pay-per-token, the cost depends heavily on the model you pick: flagship models in the Opus tier are noticeably pricier, the middle class like Sonnet is the working balance of price and quality, and light models like Haiku handle simple tasks for less. A useful habit is checking your spend after every session: agentic work with iterations burns tokens faster than plain chat, because the model reads and re-verifies a lot.

For a user in Russia both official routes hit the same wall: you need a foreign card. Paying for an Anthropic subscription or topping up an API balance with a Russian card directly will not work, and workarounds via intermediaries and virtual cards add fees and the risk of losing access at any moment.

Kumo is an OpenAI-compatible API gateway with a single balance for dozens of models. Kumo's site states that Claude Code works with a Kumo balance directly, no Anthropic subscription required: the agent connects through the gateway, and you pay only for actual tokens. Any model in the catalog is available — Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5, alongside GPT-5.6, Gemini 3 Flash, DeepSeek V4 Pro and others.

Setup does not require hand-crafting configs: the Kumo dashboard has an Integrations section with a ready-made config for Claude Code and other tools, with details in the docs. You get a key after signing up, then simply follow the Integrations instructions.

The main point for our scenario is payment. The balance is topped up with a Russian card or via SBP, with no VPN and no foreign card. There are no subscriptions, seats or minimum payments: you add funds and spend them on tokens. The effective rate is 30–50% below the labs' official price lists; the exact savings depend on your workload profile and volume, and the price for every model is visible before you spend a single token.

Spending stays under control: each API key gets limits and alerts, and you can create a separate key per project. The model is never swapped — request Claude Sonnet 5 and you get exactly that, with every response reporting which model served it. Kumo does not log prompt bodies; only billing metadata is stored. This turns Claude Code from a tool you cannot figure out how to pay for into a daily working agent with a transparent bill.