@three-ws/loom-mcp

Loom — the three.ws community 3D-creation gallery, from any AI agent. Browse the public feed of community-forged GLB creations, fetch one with its viewer URL, and submit your own.

---

A Model Context Protocol server that exposes Loom, the three.ws community 3D-creation gallery, over stdio. Browse what others have forged, pull any creation with a ready-to-embed viewer URL, and contribute your own GLB to the public feed.

Loom closes the loop with the rest of the three.ws 3D suite: forge a model with @three-ws/avatar-agent or compose a world with @three-ws/scene-mcp, then publish it here for the whole platform to see. Every read is live and public; submitting needs no key or account — it is gated server-side only by an IP rate limit.

Install

npm install @three-ws/loom-mcp

Or run with npx (no install):

npx @three-ws/loom-mcp

Quick start

Claude Code, one line:

claude mcp add loom -- npx -y @three-ws/loom-mcp

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

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

Inspect the surface with the MCP Inspector:

npx -y @modelcontextprotocol/inspector npx @three-ws/loom-mcp

Tools

Tool	Type	What it does
get_loom_feed	read-only	Browse the public gallery of community-forged 3D creations, newest-first, paginated via a nextBefore cursor.
get_creation	read-only	Fetch one creation by id, with its full metadata and an inline-previewable viewer URL + iframe.
submit_creation	write	Contribute a forged GLB to the public gallery. No key required; rate-limited (~20/hour per IP).

Each read returns the creation's glbUrl plus a viewer_url (the /forge/embed orbit + AR viewer), an og_image_url (social card), and a paste-ready iframe_snippet, so any MCP client can preview the model inline.

Input parameters

get_loom_feed — limit (1–120, default 60), before (ms-epoch cursor; pass the prior page's nextBefore to page backwards).

get_creation — id (required — the creation UUID).

submit_creation — prompt (required, ≤1000 chars), glbUrl (required — see allowed hosts below), author (optional attribution, ≤40 chars, defaults to anon), previewImageUrl, tier, backend (all optional).

Submitting a creation

submit_creation posts to the public, world-readable gallery — only submit models you intend to share. There is no account, key, or signer: the endpoint is open and gated server-side by an IP rate limit (~20 submissions/hour) plus strict input sanitization. author is free-text attribution, not an authenticated identity.

The glbUrl must be an https URL hosted on one of these domains (enforced both client-side here and server-side):

three.ws · r2.dev · cloudflarestorage.com · replicate.delivery · githubusercontent.com

Re-submitting the exact same glbUrl returns the existing creation instead of duplicating it. The dedupe is best-effort by design: it scans only the newest few entries, which is where an accidental re-POST lands. Re-submitting a GLB that has since scrolled down the feed creates a second record.

get_creation answers an id that no longer exists with a typed not_found error rather than an empty creation, so a stale id in a saved reference is unambiguous.

Example

// submit_creation
> {
    "prompt": "a cowboy hat, low-poly",
    "glbUrl": "https://three.ws/accessories/hat-cowboy.glb",
    "author": "nova",
    "backend": "hunyuan"
  }
{
  "ok": true,
  "creation": {
    "id": "f295398e-2d25-410d-a82f-c8725295a1b3",
    "prompt": "a cowboy hat, low-poly",
    "glbUrl": "https://three.ws/accessories/hat-cowboy.glb",
    "author": "nova",
    "createdAt": 1750000000000,
    "viewer_url": "https://three.ws/forge/embed?src=https%3A%2F%2Fthree.ws%2Faccessories%2Fhat-cowboy.glb&title=a+cowboy+hat%2C+low-poly",
    "og_image_url": "https://three.ws/api/avatar-og?src=https%3A%2F%2Fthree.ws%2Faccessories%2Fhat-cowboy.glb",
    "iframe_snippet": "<iframe src=\"https://three.ws/forge/embed?src=…\" width=\"640\" height=\"360\" …></iframe>"
  }
}

// get_loom_feed
> { "limit": 2 }
{ "ok": true, "count": 2, "nextBefore": 1749999000000, "has_more": true,
  "creations": [ { "id": "…", "prompt": "…", "glbUrl": "…", "viewer_url": "…", "iframe_snippet": "…" } ] }

Drop a creation's iframe_snippet onto any web page, or open its viewer_url to orbit the model (with AR on supported devices).

Examples

Runnable examples live in examples/:

node examples/list-tools.mjs     # every tool with its full input schema
node examples/browse-loom.mjs    # a page of the gallery + one creation's embed, live

Both spawn this server over stdio and read the live public gallery. They are deliberately read-only: submit_creation posts to a world-readable feed, so no example writes on your behalf. See examples/README.md for expected output.

Requirements

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

Environment variables

Variable	Required	Default
THREE_WS_BASE	no	https://three.ws
THREE_WS_TIMEOUT_MS	no	20000

No API key, signer, or account is needed for any tool.

Links

• Homepage: https://three.ws
• Forge gallery: https://three.ws/forge
• Changelog: https://three.ws/changelog
• Issues: https://github.com/nirholas/three.ws/issues
• License: Apache-2.0, see LICENSE

---

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