Doop

Doop is a collaborative design canvas built around AI agents as visible participants. Agents join through the Model Context Protocol (MCP), create frames, write real copy, respond to comments, and review their output while humans steer the work. Claude Code and Codex are named as out-of-the-box clients, and the vendor says any MCP client supporting streamable HTTP and OAuth can connect. The agent keeps running on the model or subscription the user already pays for; Doop supplies the shared canvas rather than another bundled model subscription.

Agent-native design workflow

Design streams into frames as the agent works, so collaborators can see progress rather than wait for a completed generation. A comment becomes an attributed task, the selected agent performs it, and the response can include a screenshot. Doop's built-in headless renderer gives agents screenshots of their own frames, allowing checks for clipping, spacing, and contrast before human review. Every frame can also have a live image URL that re-renders when the design changes, making embedded previews less likely to become stale.

The canvas stores tasks, decisions, and comments as shared context available to the next connected agent. Doop says repeated human feedback is distilled into a design-taste profile—for example, preferred corner radius, a constrained palette, reduced shadows, or typography choices—and applied across future frames. Users can paste reference screenshots so an agent extracts a palette, typography, and mood into a brief. Pasting a public URL imports the page as an editable snapshot while preserving the original for side-by-side variants. The built-in crew includes a generalist, UX lead, copywriter, brand-compliance reviewer, accessibility reviewer, and visual-polish specialist.

Access, hosting, and pricing

Connecting an agent uses a browser sign-in and approval flow, after which its token carries the user's identity for attribution. This is better than sharing an API key, but administrators still need token revocation, workspace permissions, audit logs, and least-privilege controls. The vendor says Doop can be self-hosted as a Docker container plus PostgreSQL, and links public source on GitHub; license obligations should be checked before commercial deployment.

Hosted Doop is advertised as free while in beta, with browser and macOS access. The conventional pricing response contained no usable plan table, and no post-beta seat, storage, export, or usage price was published. This record is flagged for manual verification. Users also remain responsible for the cost of their connected agent and model.

Honest assessment

Doop differs from prompt-to-image tools because agents manipulate a persistent, collaborative canvas and can inspect their own rendered work. It is useful for web concepts, landing-page variants, UX flows, copy review, accessibility checks, and asynchronous design feedback. It is less proven for production handoff, component systems, exact code export, offline work, or large-enterprise governance. Compare Figma, Canva, Framer AI, and the guide to MCP.

Pilot a five-frame landing page with two agents and two human reviewers. Track time to first usable frame, comments resolved correctly, accessibility defects, stale exports, unwanted style carryover, and manual rebuilding required in production tools. Test OAuth revocation, concurrent edits, import fidelity, self-review accuracy, source export, backups, and a PostgreSQL restore before storing critical design history.