three.ws Scenes

io.github.nirholas由 GitHub 帳號 nirholas 發布v0.2.2更新於 Oct 24, 2026

Compose a placed 3D diorama from one sentence, export a forged world, browse saved scenes.

發布方自己的服務:Scenes Library

已驗證STDIO僅桌面媒體與設計AI 與機器學習

概覽

AI 產生的概覽

把一句話描述變成已配置好的 3D 立體場景方案、鍛造完成的物件和匯出的 glTF 場景,並可瀏覽已儲存的世界。

可以做什麼
用一句話產生場景方案:標題、氛圍、配色、地面和已定位的物件提示,不進行鍛造也不儲存。一次呼叫完成編排、鍛造、合併的完整流程,取得成品 GLB 和檢視連結;也可把已鍛造的場景合併成一個 glTF 2.0 二進位檔。瀏覽最近或精選畫廊,並依 id 取得已儲存世界及其 GLB 網址和可環繞檢視的連結。
適用情境
適合讓助理依簡短文字提示產生 3D 場景構想或可直接開啟的世界檔案,或查詢既有的 three.ws 立體場景,而不需要瀏覽器來驅動流程。
執行需求
需要 Node.js 20 或更新版本,透過 npx 在本機執行。需能連線到 three.ws API;選用變數 THREE_WS_BASE 和 THREE_WS_TIMEOUT_MS 可覆寫端點位址和請求逾時。不需要 API 金鑰或帳號。
安裝前請注意
每次呼叫都會把提示詞送到公開的 three.ws 端點,由伺服器端執行 LLM 鏈並在其伺服器上鍛造物件。完整建置可能耗時數分鐘,鍛造失敗的物件會被略過。若把 THREE_WS_BASE 指向其他主機,提示詞會改送到該主機。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 three.ws Scenes,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

[three.ws]

@three-ws/scene-mcp

Speak 3D worlds into being: turn one sentence into a placed diorama from any AI agent.

[npm] [license] [node] [MCP Registry] [three.ws]


A Model Context Protocol server that gives any AI assistant the three.ws diorama pipeline over stdio. Describe a world in a sentence and compose_scene returns a placed plan (mood, palette, ground, and a set of single-object forge prompts), ready to build into an orbitable 3D scene. build_world collapses the whole pipeline (compose → forge every object → merge) into one call for agents with no browser to drive it. export_scene merges an already-forged diorama into one glTF 2.0 binary (every object a named, selectable node, plus ground and mood-tuned lighting), ready to open in Scene Studio. Browse saved worlds with get_scene and list_scenes.

This is the companion to the three.ws avatar one-liner: avatars give you a character; scenes give you a world. No API key, no signer, no payment: every call hits the public /api/diorama endpoint.

Install

bash
npm install @three-ws/scene-mcp

Or run with npx (no install):

bash
npx @three-ws/scene-mcp

Quick start

Claude Code, one line:

bash
claude mcp add scene -- npx -y @three-ws/scene-mcp

Claude Desktop / Cursor (claude_desktop_config.json or mcp.json):

json
{	"mcpServers": {		"scene": {			"command": "npx",			"args": ["-y", "@three-ws/scene-mcp"]		}	}}

Inspect the surface with the MCP Inspector:

bash
npx -y @modelcontextprotocol/inspector npx @three-ws/scene-mcp

Tools

ToolTypeWhat it does
compose_scenegenerativeOne sentence → a diorama plan: title, mood, palette, ground, and 2-8 placed single-object forge prompts.
build_worldgenerativeOne sentence → a fully forged, exported 3D world in one call: compose → forge every object (free lane) → merge into one GLB. No browser needed; can take a couple of minutes.
export_scenegenerativeMerge an already-forged diorama (objects with real glbUrls) into one glTF 2.0 binary (named, selectable nodes plus ground and mood-tuned lighting), ready for Scene Studio.
get_sceneread-onlyFetch a saved world by id (with GLB URLs) and its orbitable viewer URL.
list_scenesread-onlyBrowse the recent or featured diorama gallery.

compose_scene returns only the plan: no meshes are forged and nothing is saved. Forge each object via /api/forge, then either call export_scene yourself or POST /api/diorama {action:"save"} to mint a shareable permalink at https://three.ws/diorama?id=…. build_world does the compose-forge-export sequence for you in a single tool call. Both build_world and export_scene degrade gracefully on partial failure: objects that never forge are skipped and named in the response's skipped array; the rest of the world still exports.

Input parameters

compose_scene: prompt (required, 3-1024 chars: one sentence describing the world).

build_world: prompt (required, 3-1024 chars: one sentence describing the world).

export_scene: diorama (required: the diorama object returned by compose_scene/get_scene, with each object to include carrying status:"ready" and a real glbUrl).

get_scene: id (required: the diorama id returned at save time).

list_scenes: list (recent | featured, default recent), limit (1-50, default 24).

Example

jsonc
// compose_scene> { "prompt": "a lonely lighthouse on a stormy cliff at dusk" }{  "ok": true,  "diorama": {    "title": "Beacon at Dusk",    "mood": "dusk",    "ground": "stone",    "island": "craggy",    "palette": { "sky": ["#2b2350", "#c76b4a"], "ground": "#5b5560", "fog": "#3a3550", "accent": "#ffb27a" },    "objects": [      { "id": "obj-0", "label": "lighthouse", "prompt": "white-and-red striped stone lighthouse", "position": [0, 0, 0], "scale": 2.2, "rotationY": 0, "status": "pending", "glbUrl": null }    ]  },  "object_count": 1,  "viewer_base": "https://three.ws/diorama"}
jsonc
// build_world: the whole pipeline in one call> { "prompt": "a neon alley with a food cart and two streetlights" }{  "ok": true,  "diorama": { "title": "Neon Alley", "mood": "night", /* ...forged objects with real glbUrl values... */ },  "object_count": 4,  "ready_count": 4,  "glb_url": "https://storage.googleapis.com/.../scene-neon-alley.glb",  "scene_studio_url": "https://three.ws/scene?model=https%3A%2F%2Fstorage.googleapis.com%2F...%2Fscene-neon-alley.glb",  "exported_count": 4,  "skipped": []}

Requirements

  • Node.js >= 20.
  • Network access to https://three.ws (or your own THREE_WS_BASE).

Environment variables

VariableRequiredDefault
THREE_WS_BASEnohttps://three.ws
THREE_WS_TIMEOUT_MSno45000

Links


Part of the three.ws SDK suite: 3D AI agents, on-chain identity, and agent payments.
Website · Changelog · GitHub

來源:packages/scene-mcp/README.md,提交 c8f2cde

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包,因此其工具未知。

版本歷史

2
  1. v0.2.2最新Oct 10, 2026
  2. v0.2.1Sep 16, 2026

評分

已滿足適用檢查項的 50%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分

其他 Scenes 整合