How to Generate 3D Models From Claude, ChatGPT or Cursor

Yes — if the AI assistant you're using supports MCP (Model Context Protocol), you can generate a 3D model without leaving the chat. Connect Arcframe's MCP server to Claude, ChatGPT or Cursor, describe an object or point it at a product photo, and a generate_3d call returns a task you poll until a downloadable .glb mesh is ready. No separate app, no manual upload to a third-party generator, no exporting files between tools. The one catch: 3D generation is the single modality that is not available on Arcframe's free tier — it requires the Starter plan or higher, while video, image and audio generation all have free-tier options.

Why this is a real gap most MCP-connected AI tools have

Most AI-generation MCP servers that assistants like Claude and Cursor connect to are built around one modality — usually image generation or video generation. That covers "make me a picture" or "make me a video," but not "make me an asset I can drop into Blender, Unity or a 3D printing slicer." 3D generation is computationally heavier and the output format (a textured mesh, not a flat image) doesn't fit the same pipeline as text-to-image, so it's often left out entirely.

Arcframe's MCP server exposes video, image, audio and 3D as tools on the same connection, so an assistant that already generates your marketing video or voiceover can also generate the 3D asset for the same project, using the same account and the same credit balance.

Assistant	Connects via MCP	Typical built-in generation scope	3D via Arcframe MCP
Claude (Desktop/Code)	Yes	Text, code — no native image/video/3D	Yes, once Arcframe's MCP server is added
ChatGPT (with MCP/connectors)	Yes	Text, native image generation only	Yes, once Arcframe's MCP server is added
Cursor	Yes	Code generation, no native media generation	Yes, once Arcframe's MCP server is added

How the request actually works

Once the MCP server is connected, the flow is the same one you'd use for an image or video generation, just pointed at a different tool:

• You ask the assistant for a 3D model — from a text description (text-to-3d), from a single photo (image-to-3d), or from several photos of the same object taken from different angles (multi-image-to-3d).
• The assistant calls Arcframe's generate_3d tool, which returns a task ID immediately — generation is asynchronous, typically taking anywhere from under a minute to a few minutes depending on the model.
• The assistant (or you) polls the job until it completes and gets back a direct link to the mesh file.

The output is a standard .glb file — the same format most game engines, AR viewers and 3D printing tools already import, so there's no format conversion step after generation.

Which 3D models are actually available

Model choice is exposed to the assistant too, so you (or it) can pick for speed vs. detail:

Model	Best input	Relative cost	Good for
Hunyuan 3D Rapid	Text or image	Lowest	Fast drafts, iterating on shape before committing to a final pass
Trellis	Image	Low	Single clean product shots
Hunyuan 3D (image)	Image	Low-mid	General-purpose image-to-3D
Meshy v6	Image	Mid	Higher-detail single-image results
Meshy v6 Multi-image	Multiple images	Mid-high	More accurate geometry from several angles
Trellis 2	Image	Mid-high	Sharper detail than the original Trellis
Tripo3D v2.5	Image	High	Production-leaning single-image detail
Rodin (image)	Image	Highest	Maximum mesh and texture fidelity

What it's genuinely good for — and where it isn't

This is a real fit for turning product photography into a rotatable 3D preview for an e-commerce listing, generating quick game or AR asset drafts from a description without opening dedicated 3D software, or producing a printable mesh from a reference photo. It is not a fit for anything requiring a rigged, animation-ready character, precise engineering tolerances, or photoreal production texturing straight out of the box — image-to-3D generation produces a static mesh with baked textures, and results are only as clean as the input photo (a plain background and a fully visible subject work far better than a cluttered scene or a partially occluded object).

It's also worth being direct about the plan requirement: every 3D model above needs Arcframe's Starter plan or higher. That's different from video, image and audio generation, which all have working options on the free tier. If you're testing the MCP connection for the first time on a free account, video, image and audio calls will succeed — the 3D calls will not, until you upgrade.

Setting up the connection

Arcframe's MCP server is added the same way any MCP server is added to Claude, ChatGPT or Cursor — as a connector pointed at Arcframe's MCP endpoint, authenticated to your Arcframe account. Once it's connected, every tool — generate_video, generate_image, generate_audio and generate_3d — is available in the same conversation, spending the same account balance. New Arcframe accounts start with 20 one-time free credits and no card required, which is enough to test video, image and audio generation before deciding whether the 3D tier is worth adding. Full current pricing for every plan is at arcframe.ai/pricing.