Google Veo MCP Server

Google Veo

Google Veo generates high-fidelity 8-second videos (720p / 1080p / 4k) with natively generated audio from text prompts or images.

Workflow

1. Submit a generation request (text-to-video or image-to-video). Returns an operation_name.
2. Poll the operation status every 10-15 seconds. Typical time: 1-3 minutes.
3. Download from the video_uri in the completed response.

Data Model

• Operations — long-running handles with done flag, progress metadata, and final generateVideoResponse containing generatedSamples[].video.uri.

Models

• veo-3.1-generate-preview — latest, best quality, supports reference images and 4k
• veo-3.0-fast-generate-001 — lower latency, slightly reduced quality
• veo-3.0-generate-001 — previous generation, good quality

Prompting

• Include scene composition, camera movement, lighting, dialogue, and sound effects
• Use negative prompts to exclude artifacts (e.g. "blurry, text overlays, distorted faces")
• Max prompt length: 2000 characters

Gotchas

• Async only: Every generation is asynchronous — always poll the operation.
• Resolution/latency tradeoff: 4k takes significantly longer than 720p.
• Video extension (continuing a previous generation) is limited to 720p.
• Image-to-video: The input image becomes the first frame; describe the motion in the prompt.