Spline ships Hana V2 with MCP support

• Hana V2 opens its 2D canvas to coding agents: splinetool's Hana V2 announcement says designs can be generated through MCP or the in-app agent.
• Repository screens can make a round trip through Hana and back to a codebase, a workflow detailed in Spline's MCP documentation and enabled by the MCP access in splinetool's feature list.
• PBR-ready GLB and GLTF import, SVG and PDF frame export, a WebGPU canvas, and vector-editing upgrades arrive together, according to splinetool's release post.
• Spline's V2 demos stretch from an editable, prompt-built vacuum-cleaner room in splinetool's vacuum example to splinetool's hand-tracking post and splinetool's object-slicing clip.

Hana is Spline's browser-native vector design surface, as its original introduction described it. The new wrinkle is a shared local MCP connection: Spline's documentation routes UI and layout jobs to Hana, while scene, mesh, camera, and lighting requests go to Spline 3D.

Hana's agent route shares a desktop MCP server with Spline 3D. The launch thread sends readers to the Hana V2 guide for the complete feature list.

The server turns an AI client's request into structured tool calls against an open editor tab, according to Spline's MCP documentation. Hana and 3D therefore act as separate endpoints for the same agent connection.

• The server is bundled with Spline's macOS and Windows desktop app. It has no separate package, and MCP does not run from the browser.
• Spline auto-registers Claude Code, Claude Desktop, Cursor, VS Code, Antigravity, and ChatGPT when it finds them installed; other MCP clients use a generated config entry.
• Calls stay on 127.0.0.1 between the desktop app and connected client. Spline's cloud AI-generation features remain the exception.

The MCP docs describe a concrete design-to-code loop rather than a one-way mockup export. It has three stages:

1. An agent can take a repository page or component and rebuild it as an editable Hana frame by translating it into Hana's supported HTML and CSS flexbox subset, per the MCP docs.
2. It can author a Hana screen directly from HTML and CSS, creating a live canvas version of a described interface, according to the same documentation.
3. After a canvas edit, the agent can export the frame and translate it back into a framework such as Next.js or Vue; each exported element keeps a stable reference to its Hana object, Spline says.

Complex or framework-specific markup may require the agent to reconcile Hana's exported HTML with the existing components, the docs add.

Hana V2's asset path now crosses from 2D composition into editable 3D content.

• GLB and GLTF models import with PBR support, splinetool's announcement lists.
• Frames export as SVG or PDF in the same update, according to splinetool's export list.
• The canvas moves to WebGPU alongside improved vector editing, splinetool's release notes says.

Spline's separate V2 launch post describes the companion 3D rebuild's WebGPU renderer, PBR/HDR material workflow, and desktop tabs that can hold Hana files beside 3D scenes.

A featured project by @gaborpribek uses the in-editor agent or MCP to create a furnished room with a vacuum cleaner, with the room and appliance retained as editable objects, splinetool's vacuum example says.

Spline posted a hand-tracking experiment for V2 alongside a separate object-slicing study.

A dust-shader study adds real-time Dust and Scratch controls to a material, rounding out the release's small interaction and surface experiments.