Updated June 11, 2026

Stop screenshotting other people's websites

Mobbin, then a screenshot of every page, then a Figma board, then pasting JPEGs into Claude one at a time. Competitive research is a clipboard job. One command replaces the pipeline.

You're redesigning your pricing page, so you do what everyone does: open Mobbin and scroll. The screens are gorgeous — and they're the apps Mobbin picked. Your actual competitors, the four sites you're really up against, aren't in the library.

So you collect them yourself. Open competitor one, screenshot the homepage, the pricing page, the signup flow. Narrow the window to phone width and shoot everything again. Repeat for competitors two, three, and four. Paste the lot into a Figma file and spend a while arranging the grid. Two hours in, you have a competitive audit board — frozen at the moment you shot it.

Then you try to use it. You copy one screenshot back out of Figma, paste it into Claude, and type "make our hero feel like this, but our colors." One image. No link, no width, no date, none of the other thirty-nine frames. The agent squints at a JPEG of a page it could have just visited.

The job was never screenshotting. The job is collecting other people's screens into a surface you can study, point at, and hand to your agent — and every tool in that workflow does a third of it. You are the clipboard between them.

Three tools, a third of the job each

Each piece is genuinely good at what it is — and none of them talk to each other.

Mobbin is someone else's collection: beautifully curated, and bounded by the curation. It has the apps Mobbin chose — not your competitor's pricing page, not your own staging deploy, not the one agency site you keep coming back to.

The Figma board is your collection, assembled by hand. No live URLs behind the frames, no capture dates, no phone-width pairs unless you shot those too — and it starts going stale the day you finish it.

Chat-pasting is how the collection finally reaches your agent: one lossy image at a time, stripped of which page it is, what width it was laid out at, and when it was true.

What all three are missing is the same thing: the screens, the structure, and the agent in one place.

One command

Here's the turn: your agent has a browser. It can visit the site, find the main pages itself, shoot every one of them at desktop and phone width, and lay them out on a board — in about the time it takes you to screenshot one page by hand.

>drafty present stripe.comcreate a drafty canvas of stripe.com's screensgrab every stripe.com page onto a drafty canvas
Running drafty present stripe.com — 20 screens × 2 widths
stripe.com — 20 screens · 1280px + 390px · captured just now
Say it however you like — Claude runs the command, finds the site's main pages itself, and lays out the whole site at a glance.

What comes back is a site board: the site's main screens — found from its own sitemap, up to twenty of them — each captured at desktop and phone width, labeled, timestamped, with the live URL under every frame. Not a folder of PNGs. A board.

New in Drafty
Site boards
Say drafty present plus any URL. Your agent captures the site's main screens — desktop and phone width — onto a canvas anyone can annotate. Boards can even re-shoot themselves on a schedule.
an actual board — stripe.com, presented by the commandA real Drafty site board of stripe.com: labeled screens, each a desktop and phone capture, with the live URL and capture time under every frame
No mockup: the top of a real stripe.com board — twenty screens, a desktop + phone pair each, every one of them commentable.

Nobody crawled anything and nothing was installed on the site — the agent reads what the site already publishes (its robots.txt, sitemap, and homepage links), picks the screens that matter, and shoots them with the browser on your machine.

A board you can point at — and so can your agent

The difference from the Figma file is what happens when you point. Every screenshot on the board is annotatable: tap a spot, type five words. And the comment isn't a sticky note — it carries the picture, the exact point you tapped, and the live URL the frame came from.

You, on the board
stripe.com/pricing · 09:14
JKmake ours feel like this
What lands in Claude's context
“make ours feel like this”
the frame you tapped, as an imagethe exact spot — (62%, 31%) on that framestripe.com/pricing — the live page
so Claude can open the real page — its CSS, its copy — not reverse-engineer a JPEG
Tap a spot, type five words. The agent sees the frame you saw, knows the pixel you meant, and can visit the live page behind it.

So "make ours feel like this" stops being a pasted JPEG. The agent sees the same frame you do, knows the exact pixel you meant, and can open the live page behind it — read the real CSS, the real copy — instead of reverse-engineering a screenshot.

A board is honest about what it is, too: a dated snapshot, not the live site. Every frame says when it was shot, and the URL under it is there for checking what the page says now.

What people use it for

> drafty present <any url>— or however you phrase it to Claude
teardownrivals, side by side
moodboardthe sites you love
stagingyour own deploy
watchre-shoots itself
Same command, different jobs — and the watch board re-shoots its own screens on a schedule, dated every time.

Where the pixels go

The screenshots are taken by the browser on your own machine — nothing is installed on the target site, and nothing crawls beyond the site's own sitemap and the pages you name.

The board itself is a normal canvas with your visibility settings: share the link with anyone, or keep it private — owner-only, enforced server-side — which is the right setting for boarding your own staging deploy.

Get started

One paste into Claude Code. It installs the Drafty plugin if you don't have it, then boards stripe.com — swap in your competitor, your hero site, or your own staging deploy:

claude
Set me up with Drafty (drafty.im). If the plugin isn't installed, have me run /plugin marketplace add drafty-im/drafty and then /plugin install drafty@drafty-im. Once it's ready, run drafty present stripe.com — capture its main screens at desktop and phone width onto an annotatable canvas — and give me the link.

Open the board it hands back, tap the thing you want to talk about, and type your five words. The agent does the looking.

And the loop continues inward: everything Claude builds for you opens the same way — a page you can tap, a comment that carries its context, an agent that sees what you saw.

Related help