Let AI agents design for you

You already have an AI assistant you talk to every day — Claude, ChatGPT, Gemini, or a coding agent in your terminal. Popcorn Editor can hand that assistant a drafting table. Connect it once, and from then on you ask for real deliverables in plain language: "make me an A5 flyer for Saturday's market" — and a minute later there's a finished, fully editable design sitting in your workspace.

This isn't a chatbot bolted onto the editor. Your assistant connects through Popcorn's agent interface (an MCP server), authors a real Design Document, and Popcorn compiles it into a design you can open, tweak, and export like anything you built by hand. It uses your assistant's own model, so it consumes none of your Popcorn AI credits.

Connect your assistant once#

The quickest start is Claude Desktop (or Claude.ai on the web — same steps, and note that custom connectors are a paid-plan Claude feature, currently marked beta by Anthropic):

1. In Popcorn, open Settings and find Your agent access — it shows your exact connection URL, something like https://app.example.com/mcp/popcorn.
2. In Claude, go to Settings → Connectors → Add custom connector. Give it any name and paste the URL. Leave the OAuth Client ID and Secret fields under Advanced settings empty — Popcorn handles registration automatically.
3. Click Add, then Connect. You'll land on a Popcorn sign-in page, then an Authorize screen. Approve it and you're linked.

That's it — no keys to copy, and the connection can create and edit designs on your behalf. Full walkthroughs for every client live in the docs: Claude (web, Desktop, and Claude Code), ChatGPT, Gemini CLI, and everything else (Cursor, VS Code).

Make your first request#

Now just ask — the way you'd brief a designer, not a machine. Concrete beats clever:

Design an A5 flyer for the Elm Street farmers' market, every Saturday
8am–1pm. Warm and hand-made feel, big friendly headline, mention free
parking and "cash & card accepted". Make it in Popcorn.

The more real detail you give — sizes, dates, the actual wording, the mood — the less generic the first draft. If you'd hand a human designer a one-paragraph brief, hand your agent the same one.

What actually happens behind the scenes#

Your assistant doesn't paint pixels. When you ask for the flyer, a well-behaved agent:

1. Reads the format. It pulls Popcorn's design authoring guide and the Design Document schema — both published as resources on the connection — so it knows the rules before it starts.
2. Drafts a Design Document. A structured description of your flyer: the canvas size, the type, the colors, every element.
3. Validates until clean. It runs the draft through Popcorn's validate-design check, which returns precise errors, and fixes its own mistakes in a loop. Validation is free and creates nothing, so the agent can iterate as much as it needs.
4. Creates the design. Once validation passes, create-design compiles the document into a real design in your active workspace and hands back an editor link.

Popcorn validates and compiles server-side; the thinking happens in your assistant. Every request is permission-checked and scoped to your workspace — an agent can only ever see and touch what you can.

Open it and make it yours#

The agent's reply includes an editor link. Click it and the flyer opens in Popcorn Editor as a completely ordinary design — real text frames, real shapes, real images. Nothing is flattened or locked. Swap the headline font, nudge the layout, drop in your own photo, export a print PDF. The agent gets you from blank page to first draft; the editor is where you make it yours.

Iterate by asking#

You don't have to fix everything by hand. Go back to the chat and ask for changes:

Make the headline bolder, switch the palette to green and cream,
and add a second date: "also Wednesdays in July".

The agent reads the design back, revises the document, and updates it in place. Two things make this safe to lean on:

• Your edits aren't bulldozed. Elements the agent keeps hold their identity across updates, so the revision behaves like an edit, not a rebuild.
• There's always a way back. Before every agent update, Popcorn snapshots the previous state as a restorable version — look for Before agent update in the design's version history. If a revision goes sideways, restore it from version history.

Start from a template instead#

Agents can also browse Popcorn's template gallery — search it, read a template's contents, and start your design from one. Try:

Find a market or food-event flyer template in Popcorn and adapt it
for the Elm Street farmers' market, Saturdays 8am–1pm.

You get the polish of a designed starting point with your real content already in place. Premium templates show up in the agent's search too, but starting from one requires a paid plan, just like in the editor.

CLI agents and read-only tokens#

Connector UIs like Claude and ChatGPT sign in with OAuth and get read and write access. Command-line and IDE agents — Claude Code, Gemini CLI, Cursor, VS Code — typically use an agent token instead, minted in Settings → Your agent access, where you choose its scope: Read & write or Read only.

Reach for read-only when the agent's job is to look, not touch: summarizing what's in a workspace, checking designs against brand rules, pulling content for a report, or any script that runs unattended. A read-only token can list and read designs and templates but can never create or change anything — the safest default for automation you won't be watching. Tokens can carry an expiry (30, 90, or 365 days), are shown only once when created, and are revocable anytime from the same settings section. More in Tokens and security.

---

One connection, and every assistant you already use becomes a designer with access to your workspace — drafting, revising, and starting from templates while you stay in plain language. For the full picture of what's possible (all the tools, workspace targeting, rate limits, and admin controls), start with the AI agents overview.