ContextClip icon

Visual context for coding agents

ContextClip: show your agent exactly what you mean.

Capture a window or region with the controls behind it, then paste the result into Claude Code, Cursor, or Codex. Less explaining. Less guessing. Faster UI fixes.

Download the beta See how it works ↓

Wondering what your agent actually receives? See the paste formats for Claude Code, Cursor, Codex, Grok and the rest.

Where it helps

Moments when a screenshot is not enough.

Everyday situations from UI work. Each one uses what the app does today: ⌃⌥C takes the window under the pointer, ⌃⌥R takes the region you draw.

The window looks empty. Is it?

A tool window shows nothing, yet the process behind it is clearly running. A screenshot only says it is blank. A capture says what macOS actually exposes: if the tree lists the rows, labels and buttons, the interface was built and simply is not drawn, which is a rendering bug. If the tree is empty too, the data never arrived. One capture, and your agent knows which half of the problem to look at.

Suggested by a customer who spent an afternoon on exactly this in Xcode’s device window.

Which button did you mean?

You ask the agent to rename a button or fix its action. From a screenshot it guesses. With ContextClip it gets the control’s role, title and identifier, plus the buttons next to it in the same toolbar, so it searches your code for that identifier instead of inventing a new control.

Eight points off

A layout looks wrong but you cannot say by how much. Draw a rectangle over the area with ⌃⌥R: the agent receives the exact on-screen frame of every control you boxed and compares them with your constraints or padding values. Numbers, not impressions.

Why is this control disabled?

A checkbox is greyed out, a field will not take focus. The capture carries the enabled, focused and value state of every control, so the agent sees the state the app is really in, not the state you think it is in, and can trace which condition in the code produced it.

Show only what you mean

You do not want to hand your whole project or screen to a chat. Box one dialog with ⌃⌥R: the picture is cropped and the tree keeps only the controls inside the rectangle. Paste it into Claude Code, Cursor, Codex or ChatGPT Classic. ContextClip uploads nothing by itself; you decide what goes into which chat.

Ask about any app

You are stuck in an app you did not write: where is the setting, why is this option greyed out? Point at the window and paste it into your AI chat. The chat sees the window’s actual settings, switches and their states instead of guessing from pixels, and can tell you which one to change.

A bug report the developer understands the first time

You test an app, design it or manage it, and something on screen is wrong. Instead of “the second button on the right”, the capture carries the control’s exact name, role, state and position next to the screenshot. Paste it into the ticket or the chat with the developer, and nobody has to ask which button you meant.

Capture what the pixels cannot say

A screenshot shows the screen. ContextClip shows your agent which controls you mean.

The agent receives what the interface looks like and what its elements are, so you spend less time describing which button, field, or state needs attention.

01

Point or select

Capture the window under the pointer, or draw a region around the part that matters.

02

Capture both layers

ContextClip records a screenshot and the relevant accessibility tree in one local capture.

03

Paste into your agent

The clipboard is prepared for the way Claude Code, Cursor, or Codex reads context.

Two shortcuts, three capture modes

Capture the whole window or only the region that matters.

Choose Full capture, Screenshot only, or Text tree in Settings. Then use the shortcut that matches the scope of your task.

⌃⌥C

Window capture

Captures the window under the pointer, or the front window in Screenshot only mode, and prepares the active mode for pasting.

⌃⌥R

Region capture

Draw a frame around a specific area. The screenshot and accessibility context are limited to that selection.

Settings

Choose the payload

Full capture combines screenshot and text tree. Screenshot only and Text tree send just the layer you need.

Tested delivery profiles

One capture. The right paste for each agent.

Agent apps interpret the macOS clipboard differently. ContextClip prepares the combination that works for the selected target.

Claude Code profile ready
Visual Inline image The screenshot appears in the conversation.
Structure Inline AX text The readable interface summary travels with it.
Fallback Local bundle Capture files remain available on your Mac.

Where it fits

Built for the visual part of software work.

UI

Debug the right control

Point to a clipped label, incorrect state, or misplaced button without describing its location in prose.

DX

Polish generated UI

Point to the visual fix while preserving the roles, names, identifiers, and frames behind it.

QA

Share a reproducible state

Attach the screenshot and structural context from the same moment in one capture.

Documentation for the beta

From first launch to the right agent profile.

The companion documentation covers setup, permissions, capture modes, shortcuts, agent-specific delivery, local files, and troubleshooting.

Open documentation →

Getting started→Permissions→Capture modes and shortcuts→Claude Code, Cursor, and Codex→Local files→Troubleshooting→

You decide what the AI sees

Useful when it sees your screen. Risky when it sees all of it.

ContextClip shows the AI only what you choose, and hands it over only when you paste.

Only what you point at

One window, or one box you draw. Not your whole screen, not your project.

Nothing leaves your Mac on its own

ContextClip uploads nothing. The capture waits on your clipboard and in a local folder until you paste it, into the chat you choose.

Your chat, your rules

Claude Code, Cursor, Codex, ChatGPT Classic, or any chat that accepts an image or text.

No trail left behind

Only the 25 most recent captures stay on your Mac; older ones are deleted automatically.

Permissions you grant, not assume

Screen Recording and Accessibility are asked for during setup and explained there. Once granted, they work at once, with no restart.

Questions

What is ContextClip?

What is ContextClip?

ContextClip is a macOS menu bar app that captures a window or a region together with the accessibility structure behind it, then prepares both for the way Claude Code, Cursor or Codex reads context. A screenshot shows the screen; a capture also says what the controls are.

What does my agent actually receive?

A screenshot and a readable summary of the interface in it: the role, title, identifier, state and on-screen frame of the controls. Each agent reads the clipboard differently, so ContextClip prepares the combination that target expects.

Which agents does it work with?

Claude Code, Cursor, Codex and ChatGPT Classic, plus Grok, Grok Bot, Grok Build and Kiro. In a web chat the local screenshot does not travel with the paste, but the interface text and structure do. The paste formats page describes what each one receives.

How do I take a capture?

Two shortcuts: ⌃⌥C takes a window capture, ⌃⌥R lets you draw a region around exactly what matters. The capture mode in Settings decides whether ContextClip includes the screenshot, the interface structure, or both.

Can I send only the screenshot, without the structure?

Yes. Settings has three capture modes: Full capture, Screenshot only and Text tree. Screenshot only and Text tree send just the layer you need.

Does anything leave my Mac?

ContextClip uploads nothing by itself. It puts the capture on your clipboard and saves a local copy on your Mac. Nothing reaches an AI service until you paste it into the chat you choose.

Does it capture my whole screen?

No. One window, or one region you draw. With ⌃⌥R the screenshot is cropped to that region and the interface tree is limited to the controls that intersect it.

What do I need to run it?

macOS 15.6 or later, on Apple silicon and Intel. Screen Recording and Accessibility are granted in the setup assistant, with no restart.

ContextClip icon

ContextClip beta for macOS 15.6+

Less explaining. Less guessing. Faster UI fixes.

The beta runs on macOS 15.6 or later, on Apple silicon and Intel. Grant Screen Recording and Accessibility in the setup assistant — no restart — and your first capture is one shortcut away.

Download the beta Read the documentation

Version 1.1.1 · 5.4 MB · signed and notarised