Veo

HTTPv1.0.0Credential RequiredJan 22, 2026

Documentation

Guide

Vertex AI Veo MCP Server

A stateless Python MCP server for Google's Vertex AI Veo API — generate videos from text, images, first/last frames, and extend existing videos. Built with fastmcp, following the same patterns as the Google Meet, Google Business, and Zomato MCP servers.

What it does

Category	Tools
🎬 Video Generation	`generate_video_from_text`, `generate_video_from_image`, `generate_video_from_first_and_last_frame`, `extend_video`, `generate_video_with_style_reference`
🔄 Operations	`get_operation_status`, `list_available_models`

Total: 7 tools

Available Models

Key	Model ID	Status	Best for
`veo-3.1`	veo-3.1-generate-preview	Preview	Latest quality, native audio, subject reference
`veo-3.0`	veo-3.0-generate-preview	Preview	Stable preview
`veo-2.0`	veo-2.0-generate-001	GA ✅	Production workloads
`veo-2.0-exp`	veo-2.0-generate-exp	Experimental	Style reference images

Auth — credentials format

Every tool accepts a VeoAuth object:

{
  "project_id": "your-google-cloud-project-id",
  "location": "us-central1",
  "access_token": "ya29.your_oauth2_access_token"
}

Getting an access token

Option 1 — gcloud CLI (easiest for testing):

gcloud auth login
gcloud auth print-access-token

Option 2 — Service Account (recommended for production):

gcloud iam service-accounts create veo-mcp-sa
gcloud projects add-iam-policy-binding YOUR_PROJECT \
  --member="serviceAccount:veo-mcp-sa@YOUR_PROJECT.iam.gserviceaccount.com" \
  --role="roles/aiplatform.user"
gcloud iam service-accounts keys create service_account.json \
  --iam-account=veo-mcp-sa@YOUR_PROJECT.iam.gserviceaccount.com

Then get a token from the service account:

from google.oauth2 import service_account
import google.auth.transport.requests
 
creds = service_account.Credentials.from_service_account_file(
    'service_account.json',
    scopes=['https://www.googleapis.com/auth/cloud-platform']
)
creds.refresh(google.auth.transport.requests.Request())
print(creds.token)

Setup

1. Enable the Vertex AI API

In Google Cloud Console:

Go to APIs & Services → Library
Search "Vertex AI API" → Enable

2. Install dependencies

pip install -r requirements.txt

3. Run the server

python veo_mcp_server.py --transport stdio

Connect to Claude Desktop

{
  "mcpServers": {
    "veo": {
      "command": "python",
      "args": ["/absolute/path/to/veo_mcp_server.py", "--transport", "stdio"]
    }
  }
}

Tool Examples

Text to Video

{
  "auth": { "project_id": "my-project", "location": "us-central1", "access_token": "ya29.xxx" },
  "prompt": "A golden retriever running on a beach at sunset",
  "model": "veo-2.0",
  "aspect_ratio": "16:9",
  "duration_seconds": 5,
  "count": 1,
  "output_gcs_uri":

Image to Video

{
  "auth": { ... },
  "prompt": "The flowers gently sway in the breeze",
  "image_gcs_uri": "gs://my-bucket/flowers.jpg",
  "image_mime_type": "image/jpeg",
  "model": "veo-3.1",
  "duration_seconds": 5
}

First + Last Frame

{
  "auth": { ... },
  "prompt": "A smooth transition from sunrise to sunset",
  "first_frame_gcs_uri": "gs://my-bucket/sunrise.jpg",
  "last_frame_gcs_uri": "gs://my-bucket/sunset.jpg",
  "duration_seconds": 8
}

Extend Video

{
  "auth": { ... },
  "prompt": "Continue the scene, camera pans right",
  "video_gcs_uri": "gs://my-bucket/input/clip.mp4",
  "duration_seconds": 5,
  "model": "veo-3.1"
}

⚠️ Important Notes

Video generation is a long-running operation (typically 30–120 seconds) — the tools automatically poll until completion
If polling times out, use get_operation_status with the returned operation_name to check later
For output_gcs_uri — you must have a GCS bucket created in advance. If not provided, video bytes are returned inline in the response
Pricing: approximately $0.03/second of generated video on Vertex AI

Project Structure

veo-mcp/
├── veo_mcp_server.py     # Entry point
├── veo_mcp/
│   ├── tools.py          # All 7 tool definitions
│   ├── service.py        # Vertex AI API client + operation polling
│   ├── schemas.py        # VeoAuth TypedDict
│   ├── config.py         # Model registry + logging
│   ├── cli.py            # CLI args parser
│   └── __init__.py
├── requirements.txt
├── Dockerfile
├── railway.json
└── .gitignore

Connect

Get connection URL

https://gateway.mewcp.com/veo/mcp

Request Builder

import asyncio
from fastmcp import Client

# Real MewCP gateway URL for this server
GATEWAY_URL = "https://gateway.mewcp.com/veo/mcp"
API_KEY = "YOUR_API_KEY"

headers={
    "Authorization": f"Bearer {API_KEY}"
}

async def main():
    
    client = Client(GATEWAY_URL, headers=headers)
    
    async with client:
        # Test the connection by pinging the server
        await client.ping()
        
        # List available operations
        tools = await client.list_tools()
        resources = await client.list_resources()
        prompts = await client.list_prompts()

        # Execute operations
        result = await client.call_tool("example_tool", {"param": "value"})
        print(result)

asyncio.run(main())