Spline MCP Server

The Spline MCP Server connects AI tools like ChatGPT, Claude, Cursor, Codex, and Antigravity directly to Spline. Using the Model Context Protocol (MCP), any MCP-compatible client can build and edit Spline 3D scenes and Hana designs for you — creating files, generating 3D models and images, writing layouts, and running edits — and what it builds is an editable Spline file, not code you have to maintain.

Using ChatGPT or Codex? Add the Spline plugin: Spline in ChatGPT and Codex has everything you need. For every other tool, there are two ways to connect:

Hosted server

The hosted server lives at https://mcp.spline.design/mcp. Your AI tool connects to it over the internet and signs in with your Spline account; every call then runs in a Spline editor in the cloud, one per conversation. Nothing has to be open on your machine — the agent keeps building even with no Spline tab in sight.

You    ──►  AI tool          ──►  mcp.spline.design   ──►  a Spline editor in the cloud
prompt      (ChatGPT, Claude,     (sign in once)            (Spline 3D or Hana)
             Cursor, Codex, …)

Connect your AI tool

The address is the same in every client: https://mcp.spline.design/mcp. Pick yours:

claude mcp add --transport http --scope user spline https://mcp.spline.design/mcp

{
  "mcpServers": {
    "spline": {
      "url": "https://mcp.spline.design/mcp"
    }
  }
}

{
  "servers": {
    "spline": {
      "type": "http",
      "url": "https://mcp.spline.design/mcp"
    }
  }
}

Sign in

When you sign in, your AI tool opens a browser window where you log in to Spline and approve access. Some tools only do this the first time the agent calls a Spline tool. After that the agent works in your Spline account: the files it creates are saved there like any other, and it can open files you already have.

To disconnect, remove the server from your AI tool.

Using it

Just ask for what you want — no file needs to be open first.

• "Create a 3D scene with a rotating glass sphere on a pedestal and a soft studio light."
• "Design a clean pricing page with three tiers and a call-to-action."

When the agent creates or opens a file, it gives you a link to it in the Spline web editor (in ChatGPT, the file opens in the Spline panel instead). Open it in your browser to watch the design come together live, and to keep editing it yourself — on the canvas or by asking the agent for changes. Follow-up requests in the same conversation keep working on the same file.

To bring the result into your codebase, the agent can export a Hana frame's current HTML + CSS — including edits you made by hand — and adapt it into your project. For a 3D scene, ask the agent to publish it: it updates the scene's public URL, Viewer embed or code export and hands you the live link to embed. More in Working with your codebase.

What the agent can do

The hosted server offers the same tools as the desktop app, except those that read files from your computer:

In Spline 3D

• Create and open 3D files, and build or edit scenes.
• Generate 3D models and images with Spline AI.
• Inspect the scene and objects, set the camera view, and capture screenshots to check its work.
• Write the code layer over the canvas — HUDs, interfaces and game logic.
• Publish the scene to its public URL, the Spline Viewer embed or a code export, and give you the link. The agent only publishes when you ask it to.

In Hana

• Create and open Hana files, and design or rebuild frames (web pages, app screens, UI).
• Add, move, duplicate, resize, recolor and update objects.
• Generate and edit images, review a frame, and export it to HTML + CSS.

To use your own images in Hana, add them to the file in the editor, or use the desktop app, whose agent can upload files from your machine. Every tool is listed by name in the tool reference.

Usage limits

Each tool call the agent makes counts toward a daily allowance set by your Spline plan, with a short per-minute cap on every plan to stop runaway loops. Reading guides and checking your profile don't count, and a call that fails on Spline's side is given back. Ask the agent how many Spline calls you have left today: it can check your plan and today's count. When you hit a limit, the agent is told which one and when it resets — the daily allowance at midnight UTC, the per-minute cap within a minute — and where to upgrade. Upgrades apply on the next call.

Privacy & security

• Sign-in uses OAuth: your AI tool never sees your Spline password, only a token that lets it call Spline's tools on your behalf. Removing the server from your tool stops it calling them.
• The agent's calls and their results travel between your AI tool and Spline's servers, and run in a Spline editor in the cloud. The files it works on are stored in your Spline account like any other.

Troubleshooting

Desktop app

The desktop server connects your AI coding tool to the live editor tabs in the Spline desktop app, running entirely on your machine.

How it works

The desktop server bridges your AI coding tool and the live editor tabs in the Spline desktop app:

You    ──►  AI coding tool   ──►  Spline MCP Server      ──►  your open editor tab
prompt      (Claude, Cursor,      (bundled with the app)      (Spline 3D or Hana)
             Codex, …)

• You describe what you want to your AI tool (for example, "add a floating island scene").
• Your AI tool works out which Spline tools to call, and sends them as structured requests — a tool name plus arguments, not your prompt text.
• The Spline app routes each call to the right editor tab, creating a file for you if none of that kind is open.
• The tab runs the call against your live document and returns the result, and your AI tool keeps calling until the task is done.

Because the calls run inside the editor tab, your AI tool can reach the scene, the objects, materials, cameras, and Spline's AI generation features.

One server, two surfaces. The same connection drives both products. Requests are routed automatically by kind:

• 3D requests (scenes, meshes, cameras, lighting) go to your Spline 3D tab.
• 2D / Hana requests (websites, app screens, UI, layouts) go to your Hana tab.

Requirements

• The Spline desktop app (macOS or Windows) — required, since the desktop server is bundled inside it and can't run from the browser.
• An MCP-compatible AI coding tool:
  • Claude Code, Claude Desktop, Cursor, VS Code, Antigravity and ChatGPT — configured automatically.
  • Windsurf and other MCP-compatible clients — a quick one-time setup, see Connecting other clients.

Setup

Automatic setup

When you install and open the Spline desktop app, it registers the MCP server with every supported client it finds installed. No config editing required.

1. Download the Spline desktop app (macOS or Windows) and open it at least once. This step is required — the desktop server lives inside the app.
2. Restart your AI client so it picks up the new server. Most clients only load MCP servers at startup.
3. That's it — Spline's tools are now available to the agent.

You can manage this per client in Spline's MCP settings, which lists every supported client with its own toggle plus a global switch, and shows the exact setup steps for whichever one you expand.

To confirm the connection, the tools will appear in your client namespaced under Spline (for example, mcp__Spline__3d_run_code or mcp__Spline__2d_write_html).

Connecting other clients

For an MCP-compatible tool Spline doesn't auto-configure — Windsurf, for example — copy the generated server entry into that tool's MCP config. Spline's MCP settings has an Other agents panel that gives you exactly this entry to paste, with the paths already resolved for your install.

You'll find the entry the app created under the Spline key in:

• Claude Code — ~/.claude.json
• Claude Desktop —
  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

It looks like this (the exact paths are specific to your machine and install):

{
  "mcpServers": {
    "Spline": {
      "type": "stdio",
      "command": "/path/to/Spline.app/.../MacOS/Spline",
      "args": ["/path/to/spline-mcp.cjs"],
      "env": {
        "ELECTRON_RUN_AS_NODE": "1"
      }
    }
  }
}

Copy that Spline entry into the MCP configuration of the tool you want to use, then restart it. Most tools accept this same command/args/env shape; check your tool's MCP docs if its config format differs.

Using it

1. Open the Spline desktop app and keep it running.
2. In your AI client, just ask for what you want. You don't have to open a file first — if no matching file is open, the agent creates a new one and opens it in a tab.

Example prompts

• "Create a 3D scene with a rotating glass sphere on a pedestal and a soft studio light."
• "In Hana, design a clean pricing page with three tiers and a call-to-action."
• "Generate a 3D model of a low-poly tree and add it to my scene."
• "Take the dashboard screen from src/app/dashboard/page.tsx and rebuild it in Hana so I can redesign it."
• "I refined the layout in Hana — export it and update the matching component in my repo."

The agent picks the right surface automatically: design requests go to Hana, 3D requests go to Spline 3D. If you have both open, calls are routed to the most recently focused tab of the matching kind.

As the agent works, the frame or scene it's editing lights up so you can follow along in real time. You can connect several sessions at once — they all route to the same running editor.

What the agent can do

Once connected, the agent can, among other things:

In Spline 3D

• Create new 3D files and build or edit scenes.
• Generate 3D models and images with Spline AI.
• Inspect the scene and objects, run scene edits, set the camera view, and capture screenshots.
• Write the code layer over the canvas — HUDs, interfaces and game logic. See the Preview / Edit / Code tabs and AI Agent.
• Publish the scene to its public URL, the Spline Viewer embed or a code export when you ask, and give you the live link.

In Hana

• Create new Hana files and design or rebuild frames (web pages, app screens, UI).
• Add, move, duplicate, resize, recolor, and update objects.
• Generate and edit images, upload images and other assets from your machine, work with 3D shapes, apply effects, and export designs.

Every tool is listed by name in the tool reference.

MCP settings

Open Settings → MCP in the Spline app. Every supported client gets its own row with a toggle, so you can connect Claude Code but leave Cursor alone. Expand a row to see the setup steps for that specific client.

Turning a client off removes Spline's entry from that client's config and stops it being re-added; turning it back on re-registers it. Agents only load their MCP servers at startup, so Spline reminds you to restart the one you just switched on.

If you delete the Spline entry from a config by hand, Spline treats that as a deliberate opt-out and won't put it back.

Connection status

Each row shows whether that client is currently talking to Spline:

"Starts on use" is not a problem. VS Code and ChatGPT don't start the connection when they open, so send a message in the client and it connects.

There's also an MCP status indicator in the app's sidebar — click it to jump straight to these settings.

Privacy & security

• The server binds to 127.0.0.1 only — it is never reachable from the network.
• Connections are gated by an origin allowlist, so arbitrary web pages can't reach it. There are no tokens to manage.
• Your scenes and prompts stay between the Spline app and the AI client you connected. (Spline AI generation features used by the agent run in the cloud, the same as when you use them in the editor.)

Troubleshooting

Working with your codebase

Because your AI coding tool is already working inside your repository, Spline can sit right in your design-to-code loop instead of being a separate silo. This works the same with the hosted server and the desktop app.

Spline 3D

• Put a scene in your app. Ask the agent to publish the scene and it hands back the public URL, the Spline Viewer embed, or a code export for React, Next.js, Vanilla JS, three.js or react-three-fiber, then wires it into your project. Publishing again updates the same link, so your app picks up scene changes without a code change.
• Bring UI and logic into the scene. Every scene has a code layer: a full HTML, CSS and JS document over the canvas, for HUDs, forms, interfaces and game logic. The agent can write it from code in your repo, edit it in place, and read it back out to keep your source in step.
• Build from your project's content. Point the agent at what's already in your repo — brand colors, copy, a product list — and it builds or updates the scene to match.

Hana

• Bring a screen into Hana. Point the agent at a page or component in your repo, and it rebuilds that screen as a live Hana frame — it translates the markup into Hana's supported HTML + CSS (flexbox) subset. From there you can redesign it visually on the canvas.
• Build screens in Hana, in code. You can also have the agent author a frame directly from HTML + CSS, so a screen you describe in code shows up as an editable Hana design.
• Push your canvas edits back to code. After you tweak the design in Hana, the agent exports the frame's current state back to HTML + CSS and translates those changes into your framework — Next.js, Vue, plain HTML, and so on. Every exported element keeps a stable reference back to its Hana object, so the agent can map a visual change to the right place in your source.

Tool reference

Every tool the agent can call, by name, with the server that offers it. Your AI client shows the same names, so this is the list to check against when a tool seems missing. Names and tools change as Spline adds features.

Files and account

Spline 3D

Hana