Veo 3.1

You have an MCP server that lets you generate and manipulate videos with Google’s Veo 3.1 model. It supports text-to-video, style references, frame interpolation, and batch generation while providing cost estimates and token-efficient handling of assets. This guide walks you through installing, configuring, running, and using the Veo 3.1 MCP Server so you can create high-quality videos reliably.

How to use

Set up your MCP client to connect to the Veo 3.1 MCP Server and start generating videos from prompts, reference images, or by interpolating between frames. You can perform single video generation, pre-upload references for token efficiency, extend existing Veo videos, and run batch jobs with controlled concurrency. Always estimate costs before large batches and monitor generation times to plan your workflow.

How to install

Prerequisites: you need Node.js installed on your system. Ensure you have a compatible environment for running JavaScript server code.

cd veo-mcp
npm install
npm run build

Additional configuration and usage notes

The Veo 3.1 MCP Server relies on a local stdio runtime configuration. Use Node to execute the compiled server and supply your API key via environment variables. A typical runtime configuration is shown below and should be placed in your MCP client settings.

{
  "mcpServers": {
    "veo": {
      "command": "node",
      "args": ["C:\\Users\\woute\\Githubs\\MCP\\veo-mcp\\dist\\index.js"],
      "env": {
        "GEMINI_API_KEY": "your_api_key_here"
      }
    }
  }
}

Practical usage patterns

• Start with a single video to validate prompts and settings. Choose a lower resolution (720p) and a shorter duration (4–6 seconds) to iterate quickly.

• When you have reference images, upload them once and reuse the resulting file URI across multiple generations to save tokens. You can pre-upload with an image path and reuse the file URI in subsequent requests.

• For consistent branding, generate styles using a reference image and a stable seed. Use batch jobs to explore multiple seeds or prompts in parallel while controlling concurrency to avoid rate limits.

• If you need to extend a Veo-created video, provide the existing video URI and specify how many additional seconds you want to append, along with a finishing prompt.

Troubleshooting and tips

Start by ensuring your API key is valid and the Veo 3.1 service is enabled in your Google Cloud project. If you encounter rate limits, adjust the batch concurrency and add delays between requests. For long generation times, remember that higher resolutions, audio, and reference images increase processing time.

Notes on token efficiency and costs

Token efficiency is achieved by uploading references via URLs and leveraging the Files API cache. This reduces token usage significantly compared to embedding image data directly in requests. Always estimate the expected cost before running multi-video batches.

Costs depend on model quality versus speed and whether audio is included. Use the cost estimator to project expenses for a given duration, sample count, and audio preference.

Async operation flow

Video generation is asynchronous. Start a generation request, then poll for completion to retrieve the resulting video URLs. This allows you to manage multiple concurrent requests efficiently.

Security considerations

Keep your API keys secure. Do not expose GEMINI_API_KEY in client-side code or public repositories. Use environment variables and secure storage for credentials in your MCP client configuration.

Notes on supported tools and actions

The server supports tools for starting video generation, checking status, uploading reference images, extending videos, running batch generations, and estimating costs. Each tool is designed to handle common workflows for Veo 3.1 video creation and management.

Available tools

start_video_generation

Submit a video generation request with a prompt and optional references or interpolation parameters to begin producing a new video.

get_video_job

Poll or query the status of an existing video generation job and receive the result URLs when complete.

upload_image

Pre-upload and cache reference images for reuse in multiple generations, reducing token usage.

extend_video

Extend an existing Veo-generated video by adding more duration and continuing the narrative with a new prompt.

start_batch_video_generation

Submit multiple video generation jobs in a single batch with concurrency controls to optimize throughput.

estimate_veo_cost

Estimate the cost of generating videos based on model, duration, and whether audio is included.