---
number: 117
lang: en
title: "The Agent Takes the Screenshot Itself"
date: 2026-08-04
stability: volatile
review_by: 2027-08-04
author: "Stefan Wendel"
coauthor: "Claude Opus 5.5"
coauthor_id: claude-opus-5-5
---
# The Agent Takes the Screenshot Itself

**A follow-up to post 104: for anything in the browser, no key combo needed.**

2026-08-04 · Stefan Wendel · Claude Opus 5.5

---

[Post 104](/en/104) was about handing screenshots to an agent in the terminal: `Cmd+Shift+4`, a fixed folder, done. For anything in the browser there is a shorter path: the agent looks for itself.

With Playwright, as an MCP server or a small script, the agent opens a browser, loads `localhost` and takes the screenshot without you lifting a finger. It can even run through several widths, desktop and phone for example, and check the layout itself after a CSS change.

| | Screenshot by hand (post 104) | Playwright |
|---|---|---|
| Effort per image | one key combo | none |
| Works for | anything on screen | browser and localhost |
| Good for | showing a bug you see | visual checks after every change |

The two complement each other. What you are looking at right now, you send by hand. What the agent should check after its own change, it looks at itself.

**The lesson:** Give the agent eyes for the browser. Then it checks its result before you do.