How to Ask Claude About a Screenshot on Mac

The fastest way to get help with what’s on your screen used to be retyping it into a search box — an error message, a foreign-language UI, a dense chart, a wall of text. These days you can skip all of that: take a screenshot and hand it to an AI model like Claude, and it reads the image and answers you. This guide covers both ways to do it on a Mac, from the copy-paste route to one that’s built into the screenshot itself.

Why ask an AI about a screenshot at all?

A screenshot already contains the answer to whatever question you have. The problem was never taking it — it’s getting the information out. An AI that can read images closes that gap for you:

• Explain a confusing UI you’re seeing for the first time.
• Read and explain an error dialog without typing it out.
• Translate a screenshot of foreign text into your language.
• Summarize a long article, doc, or chat thread captured on screen.
• Brainstorm or draft on top of what’s visible — alt text, a caption, a next step.

Method 1: screenshot, then paste into Claude (works on any Mac)

No setup required. Take a screenshot with the built-in shortcut — ⌘⇧4 to select an area, ⌘⇧3 for the full screen — then open the Claude app (or claude.ai) and paste or attach it. Add a prompt like “explain what this screen is showing” or “what does this error mean?” and you get an answer.

It works great, but it’s a few steps every time: capture, switch apps, attach, type, wait. For the occasional question it’s fine. For anything you do more than once a week, the switching starts to feel like the old retyping problem in a new costume.

Method 2: ask while you capture, inside iShot

iShot folds the AI step into the capture flow. Take a screenshot as usual, hit the AI action, and ask anything about that shot — the image goes straight to the model and answers stream back as Markdown. No app switching, no retyping, no pasting.

You bring your own key: the AI action is powered by your Anthropic Claude API key, or any OpenAI-compatible endpoint you configure. That means your capture goes only to the model provider you’ve chosen, and the rest of the app stays on your Mac.

Want to go from an answer to a shared note? Pair it with annotation or copying text out of the shot once the model tells you what matters.

What to ask: real examples

• Explain the interface — “walk me through what this settings screen does.”
• Debug an error — “what is this error, and how do I fix it?”
• Translate — “translate this UI text to English.”
• Summarize — “give me the five key points of this chart.”
• Draft from what you see — “write alt text for this.”

Privacy: bring your own key

Because the model runs on your own API key, you decide which provider gets your image and you see the billing on your own account. No shared subscription, no reviewer in the middle. For a quick recap, see how translation works in iShot.

Frequently asked questions

Can AI read a screenshot on a Mac?

Yes. Modern vision models like Claude accept an image and answer questions about it. On a Mac you can paste a screenshot into the Claude app, or use a tool like iShot that sends the capture straight to the model.

Do I need an API key to ask Claude about a screenshot?

In the Claude app, no — it’s just a conversation. Inside iShot’s AI action, you add your own Anthropic API key (or an OpenAI-compatible endpoint), so you control the provider and see the usage on your own account.

Does a screenshot tool with AI upload my captures?

Only for what you ask. In iShot, the image is sent to the model provider you configure (your own Claude API key) when you use the AI action; the rest of the app — OCR, annotation, scroll capture, recording — runs on your Mac.