需审批 Needs approval 多Agent Multi-agent

Doop — 人和 AI Agent 一起用的开源设计画布(自带 MCP) Doop — open-source multiplayer design canvas with built-in MCP for agents

Doop — open-source multiplayer design canvas with built-in MCP for agents

Agent 在画布上直接画帧,你和它共用同一份设计记忆。 Agents sketch frames on a shared canvas with one design memory.

Doop — 人和 AI Agent 一起用的开源设计画布(自带 MCP) 预览
打开原文Open source

详细介绍Details

Doop 自称 Paper.design 的开源替代:无限画布上的帧是真实 HTML(沙箱 iframe),人在浏览器里改,Claude Code / Codex 等任意 MCP 客户端通过内置 MCP(云端 `https://doop.design/mcp`,自托管也有)连上后,可以边画边流式更新、自评截图、接评论当任务。画布带多人光标、批注、活动流和「设计记忆」(钉住范例帧、沉淀风格规则给后来的 Agent)。官网写 Free in beta;仓库 AGPL-3.0,可用 `bun run dev` 或 Docker 自托管,也可用 doop.design 托管版。助手用你已有的模型订阅连 MCP,不必再交一套 AI 密钥。适合想和编程助手在同一张设计画布上改界面、又不愿只靠「截图→粘贴→再说一遍」的人。 Doop calls itself an open-source Paper.design alternative: frames on an infinite canvas render real HTML in sandboxed iframes; people edit in the browser while Claude Code, Codex, or any MCP client joins via the built-in MCP (cloud `https://doop.design/mcp`, also self-hosted) to stream frames, self-review screenshots, and turn comments into tasks. The canvas has live cursors, comments, an activity feed, and design memory (pinned exemplars and distilled style rules for later agents). The site says free in beta; the repo is AGPL-3.0 and runs with `bun run dev` or Docker, or on doop.design hosted. Agents keep your existing model subscription—no second AI key for Doop itself. Good when you want to co-design UI on one canvas instead of screenshot-and-paste loops.

为什么推荐Why we recommend it MCP 连接、自托管、设计记忆、Free in beta 均照官网与 README。本库没有注册 Doop、没有连 MCP、没有自托管实测。 MCP connect, self-host, design memory, and free-in-beta follow the site and README. This library did not sign up, connect MCP, or self-host.

核验记录:来源、抓取与截图说明Verification log: sources, checks, and captures

主源官网 https://doop.design/ 与仓库 kgoedecke/doop;晚间 HTTP 200,gh api:AGPL-3.0、851★,2026-08-22 创建,最近推送 2026-10-09。X 线索 https://x.com/AverageAiBro/status/2109822177621262758(2026-10-13)。预览为 Playwright 实拍官网首屏,裁 3:2 WebP 自托管。Primary sources are https://doop.design/ and kgoedecke/doop; HTTP 200 this evening; gh api: AGPL-3.0, 851★, created 2026-08-22, last push 2026-10-09. X lead https://x.com/AverageAiBro/status/2109822177621262758 (2026-10-13). Preview is a Playwright capture of the home hero, cropped to a self-hosted 3:2 WebP.

上手路径How to start

审批边界Approval boundary

Agent 会在画布上改帧与导出;合并进产品代码或对外发布前,自己过一遍视觉与文案。 Agents can edit frames and exports on the canvas; review visuals and copy yourself before shipping product code or public links.

  1. 先跑一次安全范围Run a safe first pass

    先在可撤回、不外发的范围内试跑。本条未批准前:Agent 会在画布上改帧与导出;合并进产品代码或对外发布前,自己过一遍视觉与文案。。 Try it first in a reversible, non-outbound scope. Until approved: Agents can edit frames and exports on the canvas; review visuals and copy yourself before shipping product code or public links..

  2. 存 SkillSave a Skill

    把跑通的步骤存成 Skill,下次同类任务直接复用,而不是重新口述。 Save the working steps as a Skill so the next similar job reuses them instead of being re-explained.

  3. 再上 RoutineThen add a Routine

    确认输出可靠后,再考虑定时或事件触发;例行默认仍应停在待审。 After output is reliable, consider a schedule or event trigger. Routines should still stop at review by default.

相关案例Related cases