Why I Built Playas: The Tool That Doesn't Own Your AI
The story behind Playas — why I realized design tools shouldn't be in the AI business, and how MCP changes everything.
The Realization That Changed Everything
I was building yet another AI-powered prototype when my token meter ticked past $40 for the month. Not on one tool — across three. My AI code editor, my AI design tool, my AI writing assistant. Each one wrapping the same foundation models, each one marking them up 5-10x, each one locking me into their specific provider.
I was paying for GPT-4 three times over through three different middlemen.
That's when it hit me: I already pay for AI. My Claude Pro subscription. My ChatGPT Plus account. These are powerful, general-purpose AI models that I'm already paying for. Why is every design tool asking me to pay for AI again?
AI Is a Commodity. The Canvas Is the Differentiator.
Here's the uncomfortable truth that most AI startups don't want you to hear: the AI itself is a commodity. OpenAI, Anthropic, Google — they're all converging on incredibly capable models. The difference between GPT-4o, Claude Sonnet, and Gemini Pro for UI generation is marginal and shrinking.
So what's the actual value? The canvas. The rendering engine. The node tree. The responsive override system. The real-time collaboration. The Figma export. The patch-based undo/redo. The design tool itself.
Charging for tokens on top of that is like a word processor charging you per keystroke. The value is in the tool, not in the individual operations.
That's the insight behind Playas: build the best AI-native design canvas in the world, and let users bring whatever AI they already have.
Why Charging for Tokens Is Wrong
Every AI design tool on the market today runs the same playbook: wrap an API, add a markup, sell "credits." It creates terrible incentives for everyone involved.
For users, it creates token anxiety. You self-censor your prompts. Instead of iterating freely — "try it wider," "more whitespace," "actually go back to the first version" — you ration your words because each one costs money. That's the opposite of creative freedom.
For the tool builder, it means your margins are hostage to OpenAI's pricing. When API costs go up, your margins shrink. When a competitor offers cheaper tokens, users leave. You're not building a sustainable business — you're arbitraging someone else's API.
For the ecosystem, it means vendor lock-in. Each tool picks one AI provider, builds around it, and traps users in that choice. What if Claude is better at layouts but GPT is better at color palettes? Too bad — your tool chose for you.
Playas breaks all three of these patterns. You bring your own AI. We don't charge for tokens because we never touch your tokens.
The MCP Breakthrough
The missing piece was always connectivity. How do you let any AI talk to a design tool without building custom integrations for each one?
The answer is MCP — Model Context Protocol. It's an open standard that lets AI models interact with external tools through a universal interface. Think of it as the USB port for AI: one standard connection that works with everything.
Playas exposes itself as an MCP server. That means any AI client that supports MCP — Claude Desktop, ChatGPT, Cursor, VS Code with Copilot, and more — can connect to Playas and manipulate designs directly.
Open Claude. Say "Create a landing page with a hero section, feature grid, and pricing table." Claude calls Playas through MCP, generates the node tree, and you see it appear on your canvas in real-time. No copy-pasting. No intermediary. Your AI talks directly to your canvas.
And because MCP is an open standard, every new AI client that adopts it works with Playas automatically. We don't need to build integrations. The protocol handles it.
The Vision: Your AI, Your Canvas
Playas is the design layer for the AI age. We don't compete with Claude or ChatGPT — we complement them. We're the canvas, they're the brush.
This means:
Zero token limits from Playas. Use your AI as much as you want. Generate, iterate, regenerate. We don't meter your creativity.
Use the best AI for the job. Claude for complex layouts. GPT for creative copy. Gemini for rapid iteration. Switch between them freely because Playas works with all of them.
Future-proof by default. When GPT-5 launches, or Claude Opus gets faster, or a new model appears — it works with Playas on day one. No waiting for us to "add support." If it speaks MCP, it works.
Radically simple pricing. You pay for the canvas. $9/month. That's it. No token buckets, no credit packs, no "you've used 80% of your monthly AI allowance" warnings.
What's Next
Playas is live today. Connect your AI via MCP, describe a UI, and watch it appear in seconds. The node tree architecture gives you real CSS, responsive overrides, and a design that's already structured like production code.
We're not trying to be an AI company. We're a design tool that happens to work brilliantly with AI — any AI, your AI.
The design-to-code pipeline has been broken for decades. I built Playas because I believe the fix isn't better AI — it's better tools that let you use the AI you already have.