Building an Inventory Management System from Scratch with Claude Code—Zero Code Required

Background

Claude Code (CC) is an AI programming agent from Anthropic that can generate code, run commands, fix bugs, and write documentation based on conversational prompts. The author demonstrates how to build a complete product‑inventory management system using CC and DeepSeek V4 without writing a single line of code.

Installing Claude Code

Two installation methods are provided:

Native install: curl -fsSL https://claude.ai/install.sh | bash (macOS/Linux) or

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

(Windows).

npm install: npm install -g @anthropic-ai/claude-code. If access to Claude services is blocked, configure a domestic npm registry with

# Set domestic mirror
npm config set registry https://registry.npmmirror.com

.

Verification is done by running claude --version and confirming the version output.

Configuring CC Switch and Model

Because Chinese users cannot register an Anthropic account directly, a proxy tool called CC switch is used to manage model access. After installing CC switch, the author selects Claude Official as the target model and adds it to the switch.

Three models are compared:

GLM‑5.2 – best performance, comparable to Claude Opus 4.8.

Kim‑2.6 – supports ultra‑long context, comparable to Claude Opus 4.6.

DeepSeek V4 – lower cost (≈20% of the others) with good overall capability.

The author chooses DeepSeek V4 for cost‑effectiveness but notes that GLM‑5.2 can be swapped in when higher quality is needed.

Prompt Engineering

A concise prompt is crafted for CC to generate the entire project. The prompt includes:

你是一名资深全栈架构师，请从零开发一个可直接运行的商品库存管理系统（Inventory System）。

## 技术栈
后端：Java 8、Spring Boot、MyBatis Plus、MySQL 8、Maven、Lombok、EasyExcel
前端：Vue3、Vite、Element Plus、Axios、Pinia、Vue Router
部署：Docker、docker‑compose

## 第一版功能
### 1、登录（admin / admin）
### 2、首页 Dashboard（商品总量、临期商品、已过期商品）
### 3、商品管理（分页、搜索、新增、编辑、删除、Excel 导入）

## 数据库
请设计合理的 MySQL 表结构，并提供完整 SQL。

## 项目要求
- 完整可运行的代码
- 前后端 RESTful 接口
- 统一异常处理
- 提供 README、Dockerfile、docker‑compose.yml
- 最终执行 `docker compose up -d` 完成部署

The prompt is pasted into the terminal with claude, and CC returns a full project plan within a minute.

Step‑by‑Step Development

The author follows a seven‑step workflow:

Define product shape (requirements document).

Architecture design (prototype, interaction flow, system diagram).

Database design (MySQL tables).

Backend development (Spring Boot services).

Frontend development (Vue3 UI).

System deployment (Docker).

Testing and delivery.

Each step is executed by feeding the appropriate prompt to CC, pausing for confirmation, and letting the AI generate code, configuration files, and Docker scripts.

Project Structure and Deployment

The generated project is organized with separate backend and frontend directories, Dockerfiles, and a docker‑compose.yml that mounts source code and MySQL data locally. After building the images, the system runs on port 8081 (NGINX proxy adjusted from the default 80).

Testing, Bug Fixing, and Iteration

During testing, the author discovers a bug where the inventory quantity and minimum‑stock fields misbehave. By sending a corrective prompt to claude code, the AI patches the issue, and the UI updates correctly.

Additional features, such as a stock‑warning module, can be added later by continuing the conversational workflow, effectively “building with blocks.”

Cost Summary

Token usage on DeepSeek’s platform is reported as ~20 million input tokens and ~110 k output tokens, resulting in a total cost of approximately ¥1.87 (about $0.27), indicating that the AI‑driven development is inexpensive when run outside peak hours.

Conclusion

The end‑to‑end experience shows that Claude Code, combined with a suitable model and proxy tool, can generate a production‑ready inventory management system in roughly 30 minutes, with minimal manual coding, low cost, and the ability to iterate through conversational prompts.