Doop

Product Introduction

1. Definition: Doop is an open-source, multiplayer web-based design canvas and a Model Context Protocol (MCP) server specifically for design collaboration with AI agents. It functions as a real-time, shared workspace where human designers and AI models like Claude Code, Codex, and other MCP-compatible agents can co-create visual designs.
2. Core Value Proposition: Doop exists to solve the fragmented, single-player nature of AI-assisted design by providing a persistent, shared memory and live collaboration layer. Its primary value is enabling multi-agent AI design workflows where multiple AI specialists can work concurrently on a single canvas, learn from collective decisions, and maintain consistent design context, all without requiring users to pay for additional platform-specific AI tokens.

Main Features

1. Live Multiplayer Canvas: The core interface is a real-time canvas where all activity is streamed and visible. Every AI agent's design tasks, from layout generation to copywriting, render frames live as they are created. This allows for observing any agent's progress mid-task and fosters a transparent, collaborative environment. The technology stack enables WebSocket-based streaming for instant frame updates.
2. Shared Canvas Memory & Taste Profile: This is Doop's central innovation. The canvas acts as a persistent memory layer that records all tasks, decisions, feedback comments, and approved designs. It automatically distills user feedback (e.g., "rounder corners," "use only blue") into a structured design taste profile. This profile, containing rules for typography, palette, spacing, and more, is automatically applied to all new frames and is inherited by every new agent that joins the session, ensuring brand consistency.
3. Integrated Agent Workflow & Self-Review: Doop integrates with AI agents via the standardized MCP (Model Context Protocol) over HTTP with OAuth. Agents can be assigned specialized roles (e.g., Generalist, UX Lead, Copywriter). A key technical feature is the built-in, headless renderer that provides agents with screenshots of their own frames, enabling AI self-review for design fundamentals like contrast, spacing, and fit before human review.
4. Reference-Based Styling & Live Exports: Users can paste screenshots or URLs directly onto the canvas. AI agents analyze these references to extract mood, palette, and typography, creating a written brief to guide new designs in that specific style, moving beyond generic outputs. Every design frame is automatically assigned a live, public URL that re-renders whenever the underlying design changes, ensuring shared links and embeds (like Open Graph images) are never stale.

Problems Solved

1. Pain Point: The isolation and context loss in AI-assisted design. Traditional prompts to individual AI models lack persistent memory, leading to repetitive instructions, inconsistent outputs across sessions, and no record of collective decision-making.
2. Target Audience: Product Designers & UI/UX Teams seeking to scale and systemize AI collaboration; Solo Founders & Indie Hackers needing a full design team simulated by AI; Developers (especially those using Claude Code) who want to visually prototype and iterate on UI components within a shared context.
3. Use Cases: Rapidly iterating on landing page variants with a team of AI specialists; maintaining visual consistency across a suite of marketing pages by enforcing a learned taste profile; conducting a structured design critique by routing a mockup through specialized AI agents for UX, copy, and accessibility review.

Unique Advantages

1. Differentiation: Unlike solo AI image generators (Midjourney, DALL-E) or single-session design tools, Doop is built for persistent, multi-agent collaboration. It differs from traditional design tools (Figma) by being agent-native, where the workflow, memory, and review systems are built for AI, not just augmented by it. Its "bring your own agent" model avoids vendor lock-in and markup on AI usage.
2. Key Innovation: The canvas-as-a-database model. Doop's fundamental innovation is treating the design canvas not just as a visual output, but as a structured, queryable memory store for design intent, decisions, and taste. This allows any MCP-compatible AI to plug into a pre-existing, rich context and contribute meaningfully without starting from zero, enabling truly continuous and cumulative AI-human design processes.

Frequently Asked Questions (FAQ)

1. How does Doop work with AI agents like Claude? Doop functions as an MCP (Model Context Protocol) server. You connect your existing AI agent (e.g., Claude Code running locally) to Doop's canvas via a simple command (claude mcp add). The agent then uses Doop's tools to read canvas memory, render frames, and post tasks, designing "as you" on the shared workspace using its own underlying model and your existing subscription.
2. Is Doop a replacement for Figma or traditional design software? No, Doop is a complementary AI collaboration layer. It is optimized for the generative and iterative phase of design with AI agents. It excels at rapid prototyping, generating variants, and maintaining consistency through AI-driven passes. Final pixel-perfect layout and handoff might still occur in tools like Figma, which can be imported into Doop as a reference or starting point.
3. What is the "taste profile" and how is it created? The taste profile is an auto-generated set of design rules distilled from your explicit feedback and approved designs on the canvas. When you leave comments like "use softer corners" or "stick to the blue palette," Doop's system abstracts these into reusable constraints (e.g., border-radius: 10px min, primary-color: #0052FF). This profile is stored in the canvas memory and automatically injected as context for all subsequent AI-generated frames.
4. Can I use Doop for free and what are the limits? Doop is currently free while in beta. Being open-source (Apache 2.0 licensed), you can also self-host the entire platform using Docker and Postgres, giving you full control and removing any potential platform limits. The cloud beta version does not require a credit card.
5. What AI models or agents are compatible with Doop? Doop is compatible with any AI client that supports the Model Context Protocol (MCP) over a streamable HTTP transport with OAuth. This natively includes Anthropic's Claude Code and Codex. The open MCP standard means other AI assistants and custom agents can be configured to connect to your Doop canvas server, making it model-agnostic.