MCP Omni Server

This tool provides video generation (with optional embedded audio) using Google’s Gemini Omni model (gemini-omni-1.1-flash-preview) via the Vertex Interactions API. It is one of the MCP tools for Google Cloud Genmedia services, acting as an MCP server component to allow LLMs and other MCP clients to generate videos from a text prompt, optionally conditioned on input images and/or videos.

Unlike the other genmedia servers (which use the Vertex genai client), Omni is reachable only through the Interactions API, so this server calls the shared common.GenerateOmniVideo helper (which owns the Interactions client) rather than a genai.Client. The Interactions endpoint is global-only: LOCATION defaults to global and the API call is always made against the global endpoint.

MCP Tool Definitions

The server exposes the following tool:

1. omni_video_generation

• Description: Generates video (with optional embedded audio) from a text prompt, optionally conditioned on input images and/or videos, using Google’s Gemini Omni model via the Vertex Interactions API. Returns MP4(s) saved locally and/or to GCS.
• Handler: omniVideoGenerationHandler
• Parameters:
  • prompt (string, required): The text prompt describing the video to generate.
  • model (string, optional): Model to use. Can be the canonical model ID (gemini-omni-1.1-flash-preview) or a common alias (e.g. Omni, Gemini Omni 1.1 Flash). The prior gemini-omni-flash-preview remains selectable via its ID or the Gemini Omni Flash alias. See mcp-common/models.go for the supported list. Defaults to gemini-omni-1.1-flash-preview.
  • images (array of string, optional): Up to 10 input images to condition generation on. Each entry is a local file path or a gs:// URI. The MIME type is inferred from the file extension: image/png, image/jpeg, image/webp, image/gif, image/heic, image/heif.
  • videos (array of string, optional): Input videos to reference or edit. Each entry is a local file path or a gs:// URI (e.g. video/mp4, video/webm, video/quicktime).
  • sample_count (number, optional): Number of videos to generate (1-3, default 1). Clamped to the model maximum of 3.
  • temperature (number, optional): Sampling temperature, 0.0-2.0 (higher = more varied). Sent in generation_config.
  • top_p (number, optional): Nucleus sampling probability mass, 0.0-1.0. Sent in generation_config.
  • output_directory (string, optional): Local directory to save the generated video(s) to. Filenames (omni_<timestamp>_<n>.mp4) are generated automatically unless output_filename is set.
  • output_filename (string, optional): Base name for the output(s), e.g. clip.mp4. The extension is forced to the true video type and, when more than one video is generated, a _1..n suffix is inserted before the extension. Applied identically to local files and GCS objects. See Naming Outputs.
  • gcs_bucket_uri (string, optional): GCS URI prefix to store generated video(s) (e.g., your-bucket/outputs/). If not provided and GENMEDIA_BUCKET is set, gs://<GENMEDIA_BUCKET>/omni_outputs/ is used. For each uploaded video a best-effort V4 signed HTTPS URL is also returned (see OMNI_SIGNED_URL_EXPIRY_HOURS).

Authentication

This server uses Application Default Credentials (ADC) to authenticate to the Vertex Interactions API — run gcloud auth application-default login locally, or rely on the attached service account when running on Google Cloud. The Interactions API is invoked against the global location only; per-region overrides do not apply to the Interactions call itself.

Environment Variable Configuration

The tool utilizes the following environment variables:

• GOOGLE_CLOUD_PROJECT (string): Required. Your Google Cloud Project ID. The application will terminate if this is not set. Note: PROJECT_ID is also supported as a fallback.
  • Override: You can override this globally for this specific server by setting OMNI_PROJECT_ID.
• GOOGLE_CLOUD_LOCATION (string): The preferred Google Cloud location. For Omni this defaults to global and the Interactions call is always made against the global endpoint.
  • Fallback: LOCATION is also supported as a fallback for GOOGLE_CLOUD_LOCATION.
  • Override: You can override this globally for this specific server by setting OMNI_LOCATION.
• GENMEDIA_BUCKET (string): An optional default Google Cloud Storage bucket to use for GCS outputs if the gcs_bucket_uri parameter is not specified in the tool request. The path omni_outputs/ will be appended to this bucket.
  • Default: "" (empty string).
• OMNI_SIGNED_URL_EXPIRY_HOURS (number): Validity, in hours, of the V4 signed HTTPS URL returned for each uploaded video. Clamped to 168 (the V4 maximum); 0 disables signed-URL generation (the gs:// URI is still returned). Signing under ADC without a private key requires roles/iam.serviceAccountTokenCreator on the runtime service account (non-fatal if absent).
  • Default: 24
• PORT (string, for HTTP transport): The port for the HTTP server to listen on.
  • Default: "8080"

Transports Supported

• stdio (default)
• sse (Server-Sent Events)
• http (Streamable HTTP)

CORS is enabled for the HTTP transport, allowing all origins by default.

Run

Build the tool using go build or go install.

• STDIO (Default):

  ./mcp-omni-go
  # or
  ./mcp-omni-go -transport stdio

• HTTP:

  ./mcp-omni-go -transport http
  # Optionally set PORT environment variable, e.g., PORT=8084 ./mcp-omni-go -transport http

  The MCP server will be available at http://localhost:<PORT>/mcp.
• SSE (Server-Sent Events):

  ./mcp-omni-go -transport sse
  # SSE server typically runs on port 8081 by default in this configuration.

  The MCP server will be available at http://localhost:8081.

Examples

Text-to-Video (omni_video_generation)

{
  "method": "tools/call",
  "params": {
    "name": "omni_video_generation",
    "arguments": {
      "prompt": "A majestic eagle soaring over a mountain range at sunset.",
      "sample_count": 1,
      "gcs_bucket_uri": "your-gcs-bucket/omni_outputs",
      "output_directory": "./omni_videos"
    }
  }
}

Image-conditioned generation (omni_video_generation)

{
  "method": "tools/call",
  "params": {
    "name": "omni_video_generation",
    "arguments": {
      "prompt": "Animate this landscape with a gentle breeze and flowing river.",
      "images": ["gs://your-gcs-bucket/source_images/landscape.png"],
      "output_directory": "./omni_videos"
    }
  }
}