3D AI Studio MCP Server

3d-ai-studio-mcp is a remote MCP server built with ASP.NET Core and the ModelContextProtocol.AspNetCore SDK. It exposes 3D AI Studio tools over Streamable HTTP instead of stdio, so MCP clients connect by URL while you run the server as a normal web process.

Available tools:

generate_tencent_hunyuan_pro
generate_tencent_hunyuan_rapid
generate_trellis2_model
edit_tencent_hunyuan_texture
remesh_tencent_hunyuan_model
generate_gemini3_pro_image
edit_gemini3_pro_image
generate_gemini31_flash_image
edit_gemini31_flash_image
generate_gemini25_flash_image
edit_gemini25_flash_image
generate_seedream_v5_lite_image
edit_seedream_v5_lite_image
get_generation_status
get_credit_balance

Prerequisites

.NET SDK 10.0+
A 3D AI Studio API key
Available credits in your 3D AI Studio account

Configuration

Set the required API key before starting the server:

$env:THREE_D_AI_STUDIO_API_KEY = "your-api-key"

Optional overrides:

$env:THREE_D_AI_STUDIO_BASE_URL = "https://api.3daistudio.com"
$env:ASPNETCORE_URLS = "http://0.0.0.0:6281"
$env:THREE_D_AI_STUDIO_FAILURE_LOG_PATH = "logs/three-d-ai-studio-api-failures.log"

THREE_D_AI_STUDIO_API_KEY is required for live tool calls. ASPNETCORE_URLS is optional and lets you bind the remote MCP server to a custom interface or port. THREE_D_AI_STUDIO_FAILURE_LOG_PATH is optional and controls where failed upstream API calls are appended as text entries.

Build

dotnet restore
dotnet build

Run locally

dotnet run --project .

By default, the launch profile exposes the MCP server at:

http://localhost:6281

VS Code / Copilot configuration

For remote HTTP mode, Copilot does not start the server for you. Run the server separately, then point .vscode/mcp.json at the URL:

{
  "servers": {
    "3d-ai-studio-mcp": {
      "type": "http",
      "url": "http://localhost:6281"
    }
  }
}

Tool behavior

generate_tencent_hunyuan_pro

Supports prompt-to-3D, image-to-3D, and multi-view image input.
Accepts either direct base64 or data-URI image strings, or local file paths that are converted to data URIs automatically.

generate_tencent_hunyuan_rapid

Supports prompt-to-3D and image-to-3D.
Returns a task_id that should be polled with get_generation_status.

generate_trellis2_model

Supports image-to-3D generation from either a public imageUrl, a local imageFilePath, or inline base64 image data.
Exposes TRELLIS.2 resolution, texture, seed, thumbnail, and decimation controls.

edit_tencent_hunyuan_texture

Applies a new texture to an FBX model from either a text prompt or a reference image.
Supports local image files for the image-guided mode by converting them to data URIs automatically.

remesh_tencent_hunyuan_model

Calls the Tencent smart-topology endpoint for GLB or OBJ models.
Supports optional polygon type and density presets.

generate_gemini3_pro_image, generate_gemini31_flash_image, generate_gemini25_flash_image

Submit text-to-image jobs for the Gemini image models exposed by 3D AI Studio.
The server validates model-specific aspect ratios, resolutions, and image counts before calling the API.

edit_gemini3_pro_image, edit_gemini31_flash_image, edit_gemini25_flash_image

Submit image-editing jobs with one or more source images.
Accepts either inline base64/data URI images or local file paths that are converted automatically.

generate_seedream_v5_lite_image, edit_seedream_v5_lite_image

Support Seedream V5 Lite generation and editing, including image size presets, batch size, seed, and safety checker options.
Editing accepts inline image data or local file paths that are converted automatically.

get_generation_status

Polls /v1/generation-request/{task_id}/status/.
Returns generation progress, failure details, and generated asset URLs when finished.

get_credit_balance

Calls /account/user/wallet/.
Returns the current non-expired API credit balance for the authenticated account.

Notes

The MCP endpoint is exposed through ASP.NET Core with app.MapMcp().
The server starts even if the API key is missing, but tool calls will fail with a clear configuration error until THREE_D_AI_STUDIO_API_KEY is set.
Failed upstream API calls are appended to logs/three-d-ai-studio-api-failures.log by default, including request and response JSON details.
Rapid responses ultimately yield a ZIP archive containing OBJ assets.
Pro responses yield a GLB file.
For development, plain http://localhost is usually the simplest option for MCP clients.

MCP Servers