wiri

wiri

Wiri is a headless wireframing MCP server designed for an LLM operator. It stores each screen as a small semantic scene graph, lays elements out without absolute coordinates, and closes the editing loop with deterministic renders and lint findings.

Category
访问服务器

README

Wiri

Wiri is a headless wireframing MCP server designed for an LLM operator. It stores each screen as a small semantic scene graph, lays elements out without absolute coordinates, and closes the editing loop with deterministic renders and lint findings.

The v1 contract is in SPEC.md. The persisted document shape is also available as schemas/wire-document.schema.json.

Requirements

  • Node.js 20.9 or newer
  • A local directory for documents and exports

Wiri runs entirely on the local machine. It makes no network requests and sends no telemetry. Documents, checkpoint history, rendered exports, form values, and annotations are stored as plaintext; do not use production secrets as wireframe content.

Install and run

npm install
npm run build
node dist/index.js --workspace ./wireframes

The process speaks MCP over standard input/output. It writes protocol messages only to stdout and exits when its client disconnects. WIRI_WORKSPACE can set the workspace when --workspace is omitted; the current directory is the final default.

An MCP client configuration can launch the built server like this (use absolute paths in a real configuration):

{
  "mcpServers": {
    "wiri": {
      "command": "node",
      "args": [
        "/absolute/path/to/wiri/dist/index.js",
        "--workspace",
        "/absolute/path/to/wireframes"
      ]
    }
  }
}

For development, npm run dev -- --workspace ./wireframes runs the TypeScript entry point directly.

Canonical workflow

  1. Call create_document, then add one or more screens with create_screen.
  2. Call get_tree before editing a screen that has not been read in the current context.
  3. Prefer one or two atomic batch calls to build a screen. A later operation can refer to a node created by an earlier add with $0, $1, and so on.
  4. Use render for visual feedback and lint for deterministic findings.
  5. Use checkpoint before a larger revision; diff, undo, and redo make changes inspectable and reversible.
  6. Use links or render_flow to verify navigation, then export to produce PNG, SVG, or static HTML.

A compact signup flow can be built in one document with this shape:

welcome --start_btn--> email --continue_btn--> password
password --continue_btn--> verify --verify_btn--> done

Each arrow is simply a to prop on its source button. HTML export turns those links into fragment navigation between screen sections.

Layered application surfaces use constrained anchors rather than freeform coordinates. For example, a full-screen image or map placeholder can use anchor="stretch", while any number of variant="fab" buttons live in stacks anchored to screen corners. Semantic left, right, and bottom drawers are direct children of the same layer.

Storage

The server owns one active document at a time and writes successful mutations before returning:

<workspace>/
  checkout_redesign.wire.json
  checkout_redesign.history.json
  exports/
    ...

Writes use a temporary file plus atomic rename. Opening another document clears session-only undo/redo stacks; named checkpoints persist in the history file. Export files are safe to regenerate and overwrite deterministically.

Commands

npm test          # unit and integration tests
npm run typecheck # strict TypeScript check
npm run build     # emit dist/
npm start -- --workspace ./wireframes

The server intentionally exposes tools only in v1—no MCP resources or prompts. Run node dist/index.js --help for command-line options.

Limits and troubleshooting

  • Names and ids must match [a-z][a-z0-9_]{0,31}.
  • A document supports 50 screens; a screen supports 300 nodes; a batch supports 100 operations.
  • Rendered height is capped at 4000 px.
  • If a call returns NO_ACTIVE_DOCUMENT, call create_document or open_document first.
  • If a parent is rejected, call get_tree and choose root, overlay, or an existing container id.
  • A failed call and a failed batch never change the active graph or its document file.

Wiri reads schema version 1 only. Malformed files, unknown node types, invalid props, or unsupported schema versions are rejected rather than repaired silently.

License notices

Rendered output embeds Inter from @fontsource/inter, distributed under the SIL Open Font License 1.1. See THIRD_PARTY_NOTICES.md.

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选