Spline V2: AI-Powered 3D Editor with MCP Server for Claude Code and Agent Workflows

Originally published at https://tekmag.thsite.top/spline-v2-ai-powered-3d-editor-with-mcp-server-for-claude-code-and-agent-workflows/

Spline V2, released August 20, 2026, is a complete architectural rebuild of the 3D editor that enables AI agents like Claude Code, Cursor, and VS Code to directly control live 3D scenes via the Model Context Protocol (MCP). The desktop app runs a local MCP server, supports WebGPU rendering, and introduces three modes — Edit, Code, and Preview — for seamless design-to-code workflows.

On August 20, 2026, Spline released V2 — a complete architectural rebuild of its 3D editor that finally bridges the gap between design tools and AI coding Agents. For the first time, external Agents like Claude Code, Cursor, Codex, Google Antigravity, and VS Code can drive the 3D editor directly through the Model Context Protocol (MCP), turning live scenes into collaborative workspaces.

Key Takeaways

• MCP Integration: Spline V2 bundles a local MCP server that lets AI agents (Claude Code, Cursor, Codex, VS Code) directly control 3D scenes in real-time.
• Three Modes: Edit mode for visual design, Code mode for HTML/JS development, and Preview mode for testing interactions.
• WebGPU Rebuild: Complete renderer rebuild on WebGPU with WebGL fallback, delivering significant performance gains.
• Built-In AI Agent: Spline's native AI agent can build scenes from scratch using the same tools as human creators.
• Desktop-Only MCP: The MCP server runs locally on 127.0.0.1 for security; browser users get WebGPU and AI features but not agent connectivity.

The Core Innovation: MCP Server Built In

The headline feature is the Spline MCP Server, bundled directly into the rebuilt desktop application for macOS and Windows. When the desktop app launches, it scans for supported AI clients and automatically registers the local server in their configuration files. Developers then restart their chosen AI tool and prompt it normally — and the Agent begins working inside a live Spline project.

Prompt Claude Code to "create a floating island," and it translates that request into structured MCP tool calls. Spline routes those calls to an open 3D editor tab, where they execute against the live document. The Changes appear in real time, fully editable, and preserved in the undo history. If no suitable file is open, the application creates one automatically.

Crucially, the server runs locally on 127.0.0.1 and is protected by an origin allowlist that blocks arbitrary web pages. Developers choose which clients have access through Spline's MCP settings, and prompts plus scene data stay between the desktop app and the connected client. The connection never touches a cloud relay — only AI generation features (like Spline's built-in asset creation) still run in the cloud, just as they would when started manually from the editor.

What This Changes for Developer Workflows

Before V2, developers typically managed separate conversations and exported files back and forth between design software and coding environments. With MCP, agents that already understand an active codebase can now manipulate the design environment natively. There's no context-switching between a design tool conversation and a code editor — the same agent session handles both.

Spline isn't alone in opening its doors to external agents. ElevenLabs recently launched a similar MCP integration for voice-agent infrastructure, but Spline's approach is distinct: the agent controls live, editable 3D scenes rather than generating static output. The round trip between design and code is automated — scene work goes to the 3D editor, interface Changes go to Hana (Spline's UI tool), and stable references on exported elements help the agent find the right place in source code.

Three Modes: Edit, Code, Preview

The editor now offers three distinct modes, each serving a different workflow:

• Edit Mode — The traditional canvas-based 3D editor, where users manipulate objects, materials, lights, cameras, and interactivity through the scene graph.
• Code Mode — A proper code editor with syntax highlighting, Spline API completions, and autosave. A new HTML overlay layers HTML and JavaScript over the canvas, with a bridge into the scene, enabling motion, interactivity, and whole games to be built in code rather than through the visual graph.
• Preview Mode — Plays the whole thing: scene and HTML/JS together, letting developers test interactions in real time.

Code mode isn't reserved for agents. Any developer can write HTML, CSS, and JavaScript to control the 3D scene directly, using Spline's runtime API to manage state and behavior while the scene runs. This is particularly powerful for game loops, complex animations, or UI logic that's easier expressed programmatically than through node connections.

Built-In AI Agent

Spline V2 includes its own AI agent, accessible through a new sidebar tab. The agent works through the same design system as human creators: it reads the scene, runs the editor's own tools, and takes screenshots to verify its work as it goes. It can build scenes from scratch — objects, materials, lights, cameras, booleans, particles, cloners, lathes, skies, variables, and interactivity through events and states.

Where something is easier expressed in code than in the scene graph, the agent generates HTML and JavaScript and wires it into the scene. For layout, it doesn't guess coordinates — alongside procedural generators, it can hand arrangements to a constraint solver, specifying relationships (near, apart, on top of, against, facing) and letting the solver figure out positions.

Everything the agent changes appears in the undo history and syncs to anyone else working in the file. Multiple agent sessions can connect to the same editor simultaneously, though Spline's documentation doesn't yet explain how teams can attribute or review individual MCP changes before publishing a shared file.

WebGPU Rendering Rebuild

The renderer has been rebuilt from the ground up on WebGPU, now the default with WebGL as a fallback. The performance gains are significant: scenes draw with far less overhead, so heavier files stay interactive, more objects can sit on screen at once, and files reach their first frame quicker. The new lighting, reflections, and simulation features rely on WebGPU's capabilities to run in real time.

Advanced Materials and Visual Features

Materials received a full PBR pass with image-based lighting and an HDR color pipeline, so imported models retain their shading from other tools. A new reflection layer adds screen-space reflections on WebGPU, plus a Mirror toggle for true planar mirrors.

Two new sky systems are available: Physical Sky (procedural, with its own dome light and intensity dial) and HDRi Sky (using an HDR image as the environment, with an option to keep the lighting while hiding the backdrop). Height fog joins classic depth fog and responds dynamically to camera angle.

Two new material layers expand the creative palette: Dust, which accumulates dust or snow on upward-facing surfaces, and Vertex Color, for models that carry their color in the mesh geometry itself.

Improved Modeling Tools

Spline V2 introduces a new Bevel/Chamfer tool with round mode and segment control, a live-updating toolbar, and smarter edge-loop selection. Texture projection gained axis control for cylindrical and spherical modes. Sculpting received brush symmetry, and UVs now work on sculpted and editable meshes.

The Shape Blend tool was rebuilt from scratch with much higher resolution and faster performance — any mesh can now serve as a source. A new Hair System grows hair from a scalp shape or any mesh, complete with lighting, shadows, wind simulation, and clumping behavior.

Performance and Export Improvements

Booleans were rewritten from scratch and now run approximately 4× faster, reliable on geometry that previously broke them. Cloners draw far more efficiently, keeping scenes with repeated objects smooth. Additional improvements include faster ambient occlusion, faster selection, faster tessellation, better shadow performance, improved physics contacts, and a long list of memory and idle-time fixes.

Exports are smaller too. The base runtime shrank and adapts its size to what a scene actually uses, so published scenes load faster. Developers can export to formats ranging from vanilla JavaScript and Three.js to React, Next.js, and React Three Fiber — all using WebGPU by default with WebGL fallback when necessary.

Desktop App Rebuilt from Scratch

Spline rebuilt its desktop application from the ground up, replacing the old version in place on macOS and Windows. Scenes now open as real tabs, each isolated so a heavy one can't crash the entire app. Background tabs aren't throttled either, so a rendered scene keeps running while the developer models in another tab. Hana files (Spline's UI/design tool) sit in the same strip, letting users work on both editors simultaneously.

The new desktop app is also how agents connect via Spline MCP without manual setup — launch it alongside your AI tool of choice, and Spline detects it automatically. A tab indicator lights up while the agent works, providing visual feedback on agent activity.

Why This Matters for the AI Agent Ecosystem

Spline V2 represents a meaningful shift in how AI agents interact with creative tools. Rather than generating static output (images, audio, text), agents now operate inside live, editable documents — making every change reversible, inspectable, and collaborative. The MCP standard, initially designed for code assistants, is proving to be a general-purpose Protocol for agent-tool integration.

For developers building interactive 3D experiences on the web, Spline V2 removes the friction of maintaining parallel contexts between design and code. An agent that understands your codebase can now also build your 3D scenes, and you can refine everything visually afterward without losing anything.

Availability

Spline V2 is available immediately. The desktop app replaces the previous version on macOS and Windows. The browser version also receives the WebGPU renderer, AI agent, and three-mode workflow. The MCP server is desktop-only — browser users cannot connect external agents through MCP.

---

Sources: Spline Official Blog, The New Stack, Archynewsy

Frequently Asked Questions

What is Spline V2?

Spline V2 is a complete rebuild of the Spline 3D editor, released August 20, 2026. It features WebGPU rendering, an MCP server for AI agent integration, and three workflow modes: Edit, Code, and Preview.

Which AI agents can connect to Spline V2?

Spline V2's MCP server supports Claude Code, Cursor, Codex, Google Antigravity, and VS Code. The server runs locally and registers automatically with detected AI clients.

Is the MCP server available in the browser?

No. The MCP server is desktop-only (macOS and Windows). Browser users receive the WebGPU renderer, AI agent, and three-mode workflow, but cannot connect external agents via MCP.

What rendering backend does Spline V2 use?

Spline V2 uses WebGPU as the default renderer with WebGL as a fallback. The WebGPU rebuild delivers significantly better performance for complex scenes.

Can I export Spline V2 projects to code?

Yes. Exports support vanilla JavaScript, Three.js, React, Next.js, and React Three Fiber — all using WebGPU by default with WebGL fallback when necessary.

{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{"@type": "Question", "name": "What is Spline V2?", "acceptedAnswer": {"@type": "Answer", "text": "Spline V2 is a complete rebuild of the Spline 3D editor, released August 20, 2026. It features WebGPU rendering, an MCP server for AI agent integration, and three workflow modes: Edit, Code, and Preview."}},
{"@type": "Question", "name": "Which AI agents can connect to Spline V2?", "acceptedAnswer": {"@type": "Answer", "text": "Spline V2's MCP server supports Claude Code, Cursor, Codex, Google Antigravity, and VS Code. The server runs locally and registers automatically with detected AI clients."}},
{"@type": "Question", "name": "Is the MCP server available in the browser?", "acceptedAnswer": {"@type": "Answer", "text": "No. The MCP server is desktop-only (macOS and Windows). Browser users receive the WebGPU renderer, AI agent, and three-mode workflow, but cannot connect external agents via MCP."}},
{"@type": "Question", "name": "What rendering backend does Spline V2 use?", "acceptedAnswer": {"@type": "Answer", "text": "Spline V2 uses WebGPU as the default renderer with WebGL as a fallback. The WebGPU rebuild delivers significantly better performance for complex scenes."}},
{"@type": "Question", "name": "Can I export Spline V2 projects to code?", "acceptedAnswer": {"@type": "Answer", "text": "Yes. Exports support vanilla JavaScript, Three.js, React, Next.js, and React Three Fiber — all using WebGPU by default with WebGL fallback when necessary."}}
]
}

Conclusion

This article has examined the key developments, regulatory dynamics, and market implications of this topic. As the situation continues to evolve, stakeholders should monitor upcoming milestones and assess how these changes align with their strategic priorities.