How I Cloned a Paid Editor with AI in 4 Hours and Saved $98

Motivated by the desire to avoid the 98 CNY annual fee for the MDNice editor, the author fed the official MDNice URL (https://editor.mdnice.com/) and a concise prompt to a large language model, requesting a replica that supports article editing and preview within a single project while keeping front‑end and back‑end code separate.

The AI quickly generated a basic editing and preview interface. By further prompting the model to locate the paid theme files, the AI reproduced the theme with minor imperfections, achieving a satisfactory visual result.

To add cross‑platform copy functionality—converting articles into formats suitable for platforms like WeChat and Zhihu—the author refined the AI‑generated code, handling rich‑text features such as bold fonts, code blocks, and annotations. Existing Markdown files (over 200) were imported, categorized into a structured folder hierarchy, and CRUD operations for articles were implemented.

Image upload, unavailable in the free MDNice version, was enabled by creating a GitHub Personal Access Token, granting the AI permission to push images directly to a GitHub repository. The author notes that GitHub storage is less reliable than alternatives like Alibaba Cloud OSS.

The initial prototype was built in about four hours, reaching a usable state. An additional day of weekend work fixed bugs and added details, consuming roughly 1 billion tokens—about 60 CNY when using Claude + Deepseek V4. Token‑saving tricks involved tools such as CodeGraph and Ponytail, both popular GitHub projects.

Beyond cloning, the author added custom AI capabilities: an AI‑driven spell‑checking bot that pops up a dialog and lists all detected errors, and an AI‑generated table of contents that adapts its style to the selected theme (e.g., red theme yields a red UI). An auto‑publish feature was also integrated, allowing AI to generate article summaries based on user preferences.

For quality assurance, the author built an end‑to‑end (E2E) test suite where AI simulates user clicks via an agent‑browser. Over 100 UI cases that would normally require a full day of manual testing now complete in minutes. The workflow incorporates CE (Compound Engineering) for requirement analysis, solution design, and execution planning.

Finally, the author experimented with a “Loop” mode: a pool of N requirements is fed to the AI, which autonomously performs requirement analysis, design, coding, testing, and acceptance, iterating without human supervision. While effective for small tasks, larger demands remain challenging.

The article concludes with reflections on how AI lowers the barrier for replicating traditional software, questioning the future role of programmers in an era where entire applications can be cloned in hours.