Hello! Today I want to announce a small and fairly simple application, ContextClip.
It was born out of a very pressing need. When working with Claude Code, Codex, Cursor or Grok, very often you need to quickly show the agent exactly what you see on the screen: what is not working correctly, what is done wrong, which button you are looking at and what exactly you want to change.
A screenshot is often not enough for this. You start explaining with words: "this button on the right", "this block", "wrong window", "doesn't look like that", and half the time is spent simply trying to explain to the agent what exactly you mean.
The most interesting thing to me about ContextClip is how it came to be.
The idea was mine. But then almost everything, discussion of what the application should be, design, implementation, coordination of solutions and even documentation, was done by Claude Code, Codex and Cursor.
Moreover, in many ways, they themselves determined what information they needed to receive in order to better understand such a "stupid user" like me, who throws them screenshots and tries to explain in words what exactly is wrong.
The result was, in a sense, an application made by agents for agents 🤖
I will not specifically go into detail here about how it works. Better look at the documentation.
And even more interesting: just give the ContextClip page to your agent and ask it whether such a tool would be useful in your work.
I myself am interested in what it will answer.
The beta runs on macOS 15.6 or later, on Apple silicon and Intel: download it here.