Reverse Engineering + AI Agents: The Technical Truth Behind Cloning Adobe Apps in Weeks

Title

Reverse Engineering + AI Agents: The Technical Truth Behind Cloning Adobe Apps in Weeks

Meta Description

A developer turned to reverse engineering after Adobe charged early cancellation fees, using AI Coding Agents to build 7 open-source alternatives in weeks. This article breaks down how reverse engineering works, why the Reverse Engineer Anything MCP Server hit 100K stars in six days, and the real limits of AI app cloning in performance, proprietary data, and law.

Article Content

An American developer, charged early cancellation fees by Adobe twice in a row, simply abandoned the entire Adobe suite. Instead, he turned to reverse engineering techniques and, within a few weeks, released 7 free open-source alternatives covering core tools like Photoshop, Illustrator, Premiere, and Lightroom, with support for Windows, Mac, and Linux. Behind this story lies an old technique that AI Coding Agents are completely rewriting.

What Exactly Is Reverse Engineering

Reverse engineering means starting from an already-compiled piece of software, understanding what it does internally, and then rebuilding similar functionality without looking at the source code. The traditional process is grueling: compilation erases variable names, function names, and comments, leaving engineers facing thousands of unnamed functions, forced to guess their purpose one by one and rename them one by one. Just figuring out a single feature can sometimes take weeks.

The involvement of AI Coding Agents has changed the cost structure of this step. Once an Agent is connected to reverse engineering tools, it can automatically open files, read code, rename functions, and summarize how a specific feature is implemented. The repetitive manual labor that once had to be ground out by hand is now handed to the model.

MCP (Model Context Protocol) is an open protocol proposed by Anthropic in late 2024 to let AI models connect to external tools and data sources in a standardized way. You can think of an MCP Server as a "capability plugin": it exposes a number of tool functions (such as reading files, unpacking, executing commands), and AI Coding Agents discover and call these functions through the protocol, thereby breaking past the model's inherent limitation of only being able to read and write text. Reverse Engineer Anything, mentioned in the article, is exactly such an MCP Server—it packages common reverse engineering operations into a toolset callable by Agents, so different Agent clients like Claude Code, Codex, and Cursor can all connect through the same interface without each having to implement its own reverse engineering toolchain.

Reverse Engineer Anything: 100K Stars in Six Days

According to a YouTube tech channel, a project dedicated to this—the MCP Server Reverse Engineer Anything—went from under 10K stars to over 100K on GitHub in just 6 days. Installation is a single npm command, accompanied by a skill file that instructs the Agent on how to investigate an app's functionality. Once installed, Claude Code, Codex, and Cursor can all call it directly.

The example the channel gives is Notion: have the Agent study the menu that pops up when you type a slash in Notion. With this MCP Server, the Agent unpacks Notion's bundle, restores the JavaScript code, traces the execution logic through menu-related functions, and ultimately reproduces the feature in its own app. Every link in this chain—unpacking, reading code, tracing, rebuilding—is connected and completed by the Agent.

The "skill file" mentioned in the article is a lightweight configuration mechanism in this kind of Agent ecosystem: it is usually a structured Markdown document that describes in natural language the standard operating procedure, available tools, and caveats for a certain type of task—essentially an "instruction manual" for the Agent. Compared with approaches that require retraining or fine-tuning a model, skill files cost very little—changing a piece of text can adjust the Agent's behavioral strategy. In reverse engineering scenarios, the skill file guides the Agent on the order in which to unpack, which entry functions to prioritize, and how to name and archive analysis results, so that the same MCP Server exhibits a consistent and reproducible workflow across different tasks.

AI Cannot Clone Everything

The developer's Photoshop clone is still in early alpha, with over 800 open issues, and opening a 12MP image consumes about 800MB of memory. He initially promised to catch up with Adobe within a month, but later changed his tune, saying it would take months.

The reason is that Adobe's most valuable assets are not in the client. Features like AI image generation run on Adobe's own servers, and reverse engineering the client only reveals a network request—not the model behind it. There is also a large amount of proprietary data: color profiles for different lenses, Pantone color cards, and other copyrighted resources. On color management alone, the open-source community has spent nearly 30 years achieving only partial support.

ICC profiles are standard files that describe how a device maps digital color values to actual color appearance. Camera manufacturers provide dedicated profiles for different lens and sensor combinations so that photos reproduce accurate tones in post-processing software. Pantone color cards, by contrast, are a commercial spot-color standard whose color numbers and formulas are protected by copyright and trademark; software that wants to display Pantone colors accurately must obtain authorization and embed the corresponding color libraries. What these two types of resources have in common is that their value comes from long-accumulated physical measurement data and commercial licensing, not from the algorithms themselves. Therefore, even if the client code is fully reconstructed, the absence of this data still makes it impossible to achieve color performance of equivalent precision—which is also the key reason open-source alternatives have long struggled to gain a foothold in professional imaging workflows.

Legal Red Lines Are Tightening

Reverse engineering is not legal in every scenario. According to the same source, within the past week Bungie sent a letter demanding the takedown of a game remade with AI, and Activision also took down an AI reverse-engineered Call of Duty video. Beyond technical feasibility, the boundaries of licensing and infringement are an unavoidable hurdle.

For developers, the real value of this toolset may not lie in "cloning competitors," but in quickly understanding the internal structure of a complex application for interoperability development, legacy system maintenance, or learning architecture design. Treating it as a learning tool rather than a copying tool carries much less risk.

Which app's functionality would you want to study with this kind of MCP Server?

One-line Summary

AI Coding Agents dramatically lower the cost of reverse engineering, but cloning commercial software is still constrained by server-side and data barriers.

Paragraph Summary

The article tells the story of an American developer who, after being charged early cancellation fees by Adobe repeatedly, turned to reverse engineering techniques and released 7 open-source alternatives within weeks. The core point is that AI Coding Agents have changed the cost structure of reverse engineering: work that once required weeks of manually guessing function purposes and renaming them can now be done automatically by Agents. An MCP Server called Reverse Engineer Anything went from under 10K stars to over 100K in 6 days; a single npm command connects it to Claude Code, Codex, and Cursor, and it can complete the full chain of unpacking, reading code, tracing, and rebuilding. But the article also points out the limits: Adobe's AI features run on its own servers, and proprietary data such as color profiles and Pantone color cards form a hard barrier, leaving the clone still in its early stages. On the legal front, Bungie and Activision have already demanded the takedown of AI-remade content, and reverse engineering is better suited as a tool for learning and interoperability than for copying.