Clone Space MCP
Archives live web pages for offline replay with full interactivity, enabling AI agents to analyze page structure and behavior.
README
Clone Space MCP
Archive a live web page so it replays offline with the motion intact, and so an AI agent can read how the page is actually built.
Saving a page from a browser gives you a dead skeleton: the markup survives, the behaviour does
not. clone-space-mcp targets the other end — a carousel that still slides, a GSAP timeline that
still runs, a ScrollTrigger that still fires, with the network unplugged.
How it works
| Stage | What it does |
|---|---|
| capture | Drives a real headful Chromium over the live page, records a full HAR (mode: 'full', content: 'attach'), sweeps the page adaptively to trigger lazy content, and saves the original document HTML, a serialized DOM, sourcemaps, and screenshots. |
| replay | Serves the archive with routeFromHAR(har, { notFound: 'abort' }) and navigates the original URL, so the page's real JavaScript re-executes. Nothing reaches the live network. |
| extract | Runs against the replay — deterministic and re-runnable — producing a behavior graph: animations with target, trigger, timing, easing and library; computed styles; matched CSS; bound listeners; un-minified sources. |
| serve | An MCP server with progressive disclosure: a small manifest plus drill-down tools, so an agent can explore a large page without loading all of it. |
The load-bearing decision is that replay re-executes the original JavaScript. Serializing the hydrated DOM into a standalone document is the approach this project rejects: it breaks hydration and entry animations, which is exactly the fidelity being chased.
Status
Early. The operating layer and the phase plan are in place; the pipeline is not built yet.
Current work is tracked in docs/OPEN-WORK-LEDGER.md and in the
repo's GitHub issues.
Development
Bun is the package manager and the runtime — except for code that drives a browser, which runs under Node. Playwright's client does not complete its handshake under Bun; see ADR 0001 for the measurements and what was rejected.
bun install
bun run verify # lint → typecheck → test → build — the ship gate (Bun)
bun run spike # the CDP measurement harness (Node, with the fixture served by Bun)
Agents: start with CLAUDE.md.
License
Not yet chosen.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。