A Step-by-Step Claude Code Tutorial from Bilibili: From Domestic Installation to Real-World Pitfalls

Claude Code is becoming one of the most talked-about AI programming tools among Chinese developers, but most people's usage still stops at "asking a question in a chat box and copying a snippet of code." A Bilibili practical course on Claude Code, called by many viewers "the most detailed explanation," covers everything from installation and integration to MCP, skills, and configuration files, and also specifically addresses the account-banning and stable-usage concerns most important to domestic users. The information density of this session deserves a separate breakdown.

Why bother learning Claude Code specifically

At the start of the course, the instructor raises a rhetorical question: isn't Claude Code already simple, so why offer a whole course on it? His answer is that many students don't actually use it professionally—they're essentially just having chatroom-style conversations—while a lot of the specific details behind the tool are widely ignored.

This judgment isn't an exaggeration in today's development environment. The instructor notes that in his company's production environment, "hand-crafting code," or what he calls old-school programming, has become rare; development is mostly assisted by web coding tools. For domestic programmers, the common lineup roughly consists of Claude Code, Codex, Cursor, and Trae.

Among these, Cursor and Trae are relatively simple, so the course focuses on Claude Code. The instructor also gives his own selection advice: prioritize Claude Code or Codex, because in his view the code and documentation these two coding agents produce are considerably better than Cursor and Trae. He adds, however, that this is just his personal impression—if you're already very comfortable with another agent, choose according to your needs.

The six modules covered by the course

This live course is split into six segments, basically covering the main aspects of Claude Code from getting started to advanced use:

1. Installation and integration of Claude Code
2. The new mode and its corresponding built-in commands
3. Development and configuration of MCP tools
4. Usage of SQL skills
5. Subagent-related usage
6. Claude's Markdown configuration file

The logic behind these six modules progresses from "getting it running" to "using it smoothly" to "customizing it to your own engineering habits." For developers who have already installed the tool but only used basic conversation, parts two through six are where the real efficiency gap opens up—built-in commands determine the interaction rhythm, MCP determines which external systems the tool can reach, skills and Subagents determine whether repetitive tasks can be distilled into reusable capabilities, and the configuration file cements these conventions.

MCP (Model Context Protocol) is an open protocol proposed by Anthropic at the end of 2024 to standardize how AI models connect to external tools and data sources. You can think of MCP as the "USB interface" of AI applications: the model itself can only read and write context, but through an MCP Server, Claude Code can access the local file system, databases, Git repositories, third-party APIs, and other external resources. The community already has a large number of ready-made MCP Servers, and developers can also write their own according to the protocol for Claude Code to call.

Subagent refers to an independent agent instance spawned outside the main conversation, with its own context window and task goal. The main agent can delegate a well-scoped subtask (for example, "research how a certain dependency library is used" or "batch-refactor the functions under a certain directory") to a Subagent, and only pass the conclusion back to the main thread. The benefit is that it avoids polluting the main context with large amounts of intermediate process, saves tokens, and lets different subtasks proceed in parallel.

Claude's Markdown configuration file usually refers to a file such as CLAUDE.md in the project root, used to persistently declare project conventions to the model: tech stack, directory structure, code style, common commands, prohibited items, and so on. It's essentially an "onboarding manual" for Claude Code, automatically loaded each session, reducing the cost of repeatedly explaining background.

The root cause of account bans and the idea of bypassing login

One issue singled out in the course is "how to use Claude Code stably." The instructor observes that many students get their accounts banned after a while. His judgment on the cause is that accounts get banned by Claude's official website due to its content moderation tendencies, making bans easy to trigger.

To address this, he shares a way to bypass login—no account registration is needed at login, so account bans are no longer a concern. According to him, from the first day of installing Claude Code until now, his account has never been banned. During use, you can connect either foreign models or domestic models, as the user decides.

What needs to be viewed objectively is that this is a single-source experience share; the course does not elaborate on the specific technical implementation path, nor does it provide verifiable evidence of long-term stability. For domestic developers, the compliance and sustainability of such approaches need to be assessed on their own; the more prudent approach is still to configure the toolchain so it can switch between different model backends, reducing dependence on a single account.

Who this tutorial is for

Judging from how the content is organized, this course is positioned as "filling in the gaps": if you can already run basic conversations with Claude Code but haven't systematically reviewed built-in commands, MCP configuration, and skill distillation, the six-module framework can serve as a self-check list; if you're a newcomer to AI programming tools, the installation and integration section can help you get over the first hurdle, but the advanced content afterward needs to be compared against your own usage as you go.

The instructor also explains at the beginning the nature of this content—those familiar with it can use it to fill gaps, and those unfamiliar should listen carefully. This positioning is actually quite honest: the value of tools like Claude Code isn't in "whether you can install it," but in "how deeply you use it."

One-line Summary

A breakdown of a Claude Code practical course: six major modules, MCP and skill usage, and ideas for handling account bans.

Paragraph Summary

The article breaks down a Bilibili Claude Code practical course called "the most detailed explanation." The course opens by pointing out that most people still use Claude Code as chatroom-style conversation, ignoring the details behind the tool. The instructor believes hand-written code in production environments has greatly decreased and recommends prioritizing Claude Code or Codex, while emphasizing this is only his personal impression. The course is divided into six modules: installation and integration, built-in commands, MCP tool development and configuration, SQL skills, Subagent, and Markdown configuration files, logically progressing from "getting it running" to "customizing to engineering habits." Regarding the account-ban concerns of domestic users, the instructor attributes the cause to the official website's account review mechanism and proposes the idea of bypassing login without registering an account, but the article cautions that this is single-source experience and its compliance and sustainability need to be assessed independently.