# 3dCMS: build, edit and publish real 3D websites, by hand or with AI 3dCMS is a browser editor and a Model Context Protocol (MCP) server for real 3D websites. An AI assistant connected to it can create scenes, place pages, text, images, links, models and lights in space, light and frame them, review the design, and publish the result to a public URL that search engines and other AI assistants can read. Free to start, no credit card, no code. The browser editor runs at https://3dcms.info/editor and a no-account sandbox at https://3dcms.info/try. A published scene is a real website: every page serves a complete semantic HTML document at the same URL as the 3D view, with schema.org data, real links, a plain-text twin at /text/, a sitemap and an llms.txt. There is no bot-specific version of any page. ## The MCP server Hosted connector URL: https://3dcms.info/mcp (transport: streamable HTTP; authorization: OAuth 2.1, sign in with Google). Official MCP registry name: info.3dcms/mcp. Also listed on Glama and Smithery. Discovery documents: https://3dcms.info/.well-known/oauth-protected-resource/mcp and https://3dcms.info/.well-known/oauth-authorization-server/mcp. The open local server (the mcp-server folder of the repository) runs on Node 18+ with an API token from the account page and adds generate_3d, generate_material and get_generation when a fal.ai key is set. ### Connecting #### Claude (claude.ai and Claude Desktop) 1. Open 3dcms.info/claude: it takes you straight to 3dCMS in Claude's Connectors Directory. 2. Click Connect. 3. Sign in with Google when Claude asks. That is the whole setup. Note: Or add it by URL as a custom connector: https://3dcms.info/mcp (custom connectors are a paid Claude plan feature). #### Claude Code 1. Run: claude mcp add --transport http 3dcms https://3dcms.info/mcp 2. Sign in with Google in the browser window it opens. Note: Works in the terminal and in the IDE extensions. #### ChatGPT 1. Turn on developer mode under Settings, then Connectors, then Advanced. 2. Create a connector with the MCP server URL https://3dcms.info/mcp. 3. Sign in with Google when ChatGPT asks. Note: Custom MCP connectors need developer mode on a paid ChatGPT plan. #### Gemini (the Gemini app) 1. In the Gemini app on the web, open Settings, then Personal Intelligence, then Connected Apps, and choose Add a custom app. 2. Paste the URL https://3dcms.info/mcp, sign in with Google, and click Allow. 3. In a chat, type @ and pick 3dCMS from the list. Note: Needs a personal Google account, in the US, 18 or older, with Keep Activity on. Works on the Google AI Plus plan. #### Cursor, VS Code, Windsurf, Codex CLI, Antigravity, Gemini CLI and other MCP clients 1. Add a server entry with "url": "https://3dcms.info/mcp" (transport: streamable HTTP, OAuth). 2. Sign in with Google on first use. Note: Any client that speaks streamable HTTP with OAuth works. Google is replacing Gemini CLI with Antigravity CLI for free users (since June 18, 2026); both take the same url entry. The sign-in only hands authorization to approved clients and to programs on your own machine. ### Tools - list_scenes: List the 3D scenes saved in your account. - get_scene: Read one scene: its nodes, transforms and material library. - create_scene: Create a new scene: empty, or a copy of one of yours (copyFrom). - apply_ops: Edit a scene: add, move, restyle, delete and link nodes. - list_assets: Search the images, 3D models and sounds in your asset library and the shared public library. - upload_asset: Put a file (a model, an image, a sound) into your asset library from a URL or its bytes, or copy in a shared library asset. - compose_surface: Build a printed surface (a canvas, poster, sign or label): a texture set plus text, decals, stains and spatter, saved as a panel model or a material. - measure: Measure a scene: sizes, the opening through a frame or doorway, distances between things and the clearance around them. - get_scene_screenshot: Render a scene headlessly and return the image. - probe_lighting: What is lighting a point: every lamp in reach, the sun or sky, and the ambient term, named. - probe_framing: Is a camera a good photograph: its subject on the thirds, the horizon, eye height, the lens, the way out, at desktop and phone shapes, with one suggested pose. - design_review: A designer's critique of a scene against its brief: palette, contrast, hierarchy, the arrival shots, wayfinding, light and motion, with the five fixes that matter most. - publish_scene: Publish a scene to a public, crawlable URL. ### What to say - "list my scenes": see what is already in your account. - "make a scene called Studio with a dark backdrop, a lit stage and three product cards": a new scene, built from one sentence. - "in Studio, move the second card behind the stage and make it warmer": edit by naming things, no ids needed. - "review the design of Studio": a designer's critique with the five fixes that matter most. - "publish Studio": a public, shareable 3D link (the assistant asks first). Every result includes an editorUrl that opens the scene in the browser editor, signed in. Scenes use metres, Y up, ground at y = 0, colours as #rrggbb. One apply_ops call is atomic and checkpointed; the owner can roll back in the editor. publish_scene and any paid generation ask the person first. ### Questions people ask **Can Claude, ChatGPT or Gemini build a 3D website?** Yes. Connect the assistant to the 3dCMS MCP server at https://3dcms.info/mcp and describe the site. It creates the scene, places the pages, text, images, links and models, lights and frames them, and publishes to a public URL when you say so. You can keep editing in the browser editor afterwards, and the assistant sees those edits too. **Is there an MCP server for building 3D websites?** Yes: 3dCMS is listed in the official MCP registry as info.3dcms/mcp and on the Glama and Smithery directories. The hosted connector needs nothing installed; the open local server in the mcp-server folder adds paid model and material generation. **Do I need to write code?** No. The assistant writes scene operations, not code, and the browser editor is a direct-manipulation tool with gizmos, an outliner and an inspector. Developers can also drive everything through the MCP tools or the JSON scene format. **Does a 3D site show up in Google?** Every published page serves a complete semantic HTML document at the same URL as the 3D view, with schema.org data, real links, a plain-text twin at /text/, a sitemap and an llms.txt. Search engines and AI search tools read exactly what a visitor sees. There is no bot-specific version of the page. **What does it cost?** Free to start, no credit card. Building through Claude, ChatGPT or Gemini runs on that app's own plan, and 3dCMS charges nothing for it. The Assistant inside the editor runs on your own Anthropic API key, and generating models and materials uses your own fal.ai key. Paid plans add published sites, storage, heavier scenes and your own branding. **Which tools does the MCP server expose?** Thirteen hosted tools: list_scenes, get_scene, create_scene, apply_ops, list_assets, upload_asset, compose_surface, measure, get_scene_screenshot, probe_lighting, probe_framing, design_review and publish_scene. The local server adds generate_3d, generate_material and get_generation. **Who is this for?** Freelance 3D designers and small studios who want a client-facing site they can walk through, agencies pitching spatial web design, developers evaluating MCP, and anyone who has asked an AI assistant how to make a 3D website without building a Three.js codebase. ## Where to read more - https://3dcms.info/ai - https://3dcms.info/learn/a/mcp-intro.html - https://3dcms.info/learn/a/ - https://3dcms.info/explore - https://3dcms.info/templates - https://3dcms.info/pricing