AI Tool Review: Paper.Design

Why I’m Reviewing AI Tools

I spend most of my working hours inside AI tools. Not evaluating them as a hobby. Using them to produce real deliverables for real clients spending real money. That means I find out fast what actually works in production versus what looks good in a demo video.

There’s a growing gap between the tools people talk about online and the tools that survive contact with a real workflow. Most reviews are written by people who spent 30 minutes clicking around a free trial. The Tool Bench is different. Every tool I review here is one I’ve used in actual production work, with real stakes, and I’m going to tell you exactly where it held up and where it didn’t.

The Tool

Paper.design is a design application built on a simple but powerful premise: your canvas runs on HTML and CSS, not a proprietary rendering engine. Everything you see on the canvas is code underneath (not unique to Paper– this is how Figma works). Everything you build in code can appear on the canvas. The company positions itself as “the connected canvas for teams shipping with agents,” which is a mouthful, but it describes the product well.

The feature set: a visual design canvas with real-time GPU-accelerated shaders, AI image generation, background removal, color palette extraction, and a library of animated visual effects. It runs on macOS, Windows, and web. The free tier gets you started. Pro is $20/month and unlocks essentially unlimited MCP calls (1 million per week versus 100 on free).

The competitive landscape here is worth understanding.

• Figma is the incumbent for professional design teams.
• Canva is the go-to for non-designers who need to produce marketing assets quickly.
• Pencil is doing something very similar to Paper is a slighly more flexible way, often better for less technical users that may get a "feature overload" feeling from Paper.design.
• Framer occupies a middle ground for code-aware designers building websites.

Paper is doing something none of them do: building the entire canvas on top of web standards so AI agents can read and write designs as code, natively, without conversion layers, including in enterprise teams.

Paper exposes 24 tools through an authenticated MCP server. Read tools let agents inspect selections, pull JSX, grab screenshots, and read computed styles. Write tools let agents create artboards, write HTML directly to the canvas, set text content, and update styles. That bidirectional access is not a gimmick. It changes the workflow fundamentally.

Bottom Line Up Front

Conditional recommend. Paper is the best tool I’ve found for one specific workflow: building design assets in a coding environment (Claude Code, Cursor) and pushing them into a visual canvas for review and refinement without losing fidelity. If you work in code and need a visual design environment that speaks the same language, Paper solves a real problem that nothing else solves well.

An important note: To get the most out of Paper or similar tools like Pencil, you must first learn to develop and store your site's "design language" in a language that the LLMs can understand. I wrote a piece about this a few weeks back that teaches you how to do that. Subscribers can download the Creative Director suite to run the workflow in your Claude Cowork or Cursor space.

Paper is for the growing number of people who live in an IDE or terminal and need their design environment to meet them where they are. If you’re a traditional designer who works spatially in Figma, Paper will feel unfamiliar and probably unnecessary. If you’re a non-technical marketer who needs to make social graphics in 10 minutes, Canva is still your tool.

What I Liked

The code-to-canvas bridge is real. When I’m working in Claude Code, design elements live in JSON files, SVG markup, HTML, and CSS. Getting those elements into a traditional design tool like Figma means saving them as vector graphics, importing them manually, and losing editability in translation. Paper eliminates that entire friction layer. I push coded elements directly into Paper through MCP, and they render exactly as they would in a browser. No export/import dance. No flattened layers. No lost context.

MCP integration is deep, not decorative. A lot of tools have announced “AI integrations” that amount to a chat sidebar bolted onto the existing product. Paper’s 24-point MCP server is different. Agents can inspect what’s on the canvas, modify it, create new elements, and read computed styles back. The push-pull loop between my CLI and the visual canvas is tight. I can build in code, review visually, make adjustments in either environment, and stay in sync.

Free tier is genuinely usable. 100 MCP calls per week on the free plan is enough to evaluate whether the tool fits your workflow before you spend anything. The $20/month Pro plan is reasonable for what you get. Compare that to Figma’s professional tier at $15/seat/month, which doesn’t include any of this agent-native functionality.

It respects the actual workflow. Paper doesn’t try to make designers write code or make developers become designers. It creates a shared surface where both representations coexist. That’s the right abstraction. The canvas shows you what the code produces. The code defines what the canvas contains. One source of truth, two views.

What I Disliked

The learning curve is real. If you’re coming from Canva, where everything is drag-and-drop with templates, Paper will feel like landing on a different planet. The interface assumes familiarity with web development concepts. You need to understand HTML structure, CSS properties, and how components nest to get real value from the tool. This isn’t a criticism of the product so much as a clear boundary on who it’s for.

Enterprise DNA shows in unexpected places. Paper is clearly building for teams at scale, which means some features feel heavy when you’re a solo operator or small agency. The workspace and collaboration infrastructure is more robust than I need for my use case. Not a dealbreaker, but you’ll notice scaffolding designed for 50-person design teams when you’re working alone.

The ecosystem is still early. Figma has a decade of plugins, community files, and integrations. Canva has a massive template library. Paper’s ecosystem is nascent. You’re betting on a trajectory, not a mature platform. The core product is strong, but if you need a specific integration or workflow that doesn’t exist yet, you’re waiting or building it yourself.

Documentation could be deeper. The MCP integration is powerful, but I found myself discovering capabilities through experimentation rather than documentation. For a tool that’s positioning itself as agent-native, the guides for setting up and optimizing agent workflows could be more thorough.

What to Do

If Paper sounds like it fits your workflow, download the free desktop app, connect it to your Claude or Cursor MCP configuration, and try pushing a real design element from your coding environment to the canvas. You’ll know within 30 minutes whether this solves a problem you actually have. Don't forget to build your LLM-friendly site design language ahead of time for the maximum quality output.

If you’re not working in code at all, Paper isn’t the right starting point. Canva remains the fastest path from idea to finished graphic for non-technical users. Figma remains the standard for professional design teams working in traditional visual workflows. For bulk ad creative, I'm still figuring out the role Paper can play. I like Templated.io for this use case, and am excited to try out Figma's new "slots" feature. You can read more about that in my Creative Asset Production series.

If you need code-to-design bridging but want something more website-focused, look at Framer. It occupies a different niche (website builder with code awareness) but shares some of Paper’s philosophy around code and design living together.

And if evaluating all of this feels like a lot, this is what we do at Marketer in the Loop. Reply to any MITL email or message me on LinkedIn– I'd love to jam on this with you!