Business & product

AI writes the JSON, Zvid renders it.

This is not text-to-video. An AI agent authors a structured JSON project through the MCP server, and Zvid renders it deterministically with FFmpeg — so the same document always produces the same frame-accurate result.

Agent-authored

AI plans the video, FFmpeg renders it

"AI-generated video" with Zvid does not mean diffusion or generated pixels. It means an AI agent — Claude Code, Cursor, Codex, or any LLM — authors a JSON project describing scenes, elements, transitions, and captions, then Zvid renders that document with FFmpeg and returns a hosted cdn.zvid.io URL. Because the engine is deterministic, the same JSON always produces the same frames.

The agent connects through Zvid's MCP server, which exposes 30 tools plus schema-aware authoring so the model builds valid projects instead of guessing at shape. Before a single credit is spent, a free pre-flight validator resolves templates and applies your plan limits, and a repair tool fixes malformed projects. You get output an agent and your brand can both rely on.

What you get

What agent-authored video gives you

30 tools for agents

Zvid's MCP server exposes 30 tools to AI agents like Claude Code, Claude Desktop, Codex, and Cursor, so the model authors valid JSON projects instead of freehanding structure.

Free pre-flight validation

The validator resolves templates and applies your plan limits without spending any credits, so an agent knows a project will render before you commit to it.

Deterministic FFmpeg output

Rendering runs on FFmpeg, not a headless browser, so the same JSON produces the same frame-accurate result every time — reproducible enough to trust in production.

Self-repairing projects

A dedicated repair tool fixes malformed JSON, and best-effort bulk validation returns invalid items in itemErrors by index while the valid ones keep rendering.

Schema-aware authoring

The MCP integration is schema-aware, guiding the agent to the correct field shapes across scenes, elements, transitions, and 12 caption modes as it builds the document.

Get pinged on completion

HMAC-SHA256 signed webhooks fire render.completed and render.failed with 5 retries, so an agent-driven pipeline reacts the moment a render lands.

30

MCP agent tools

FREE

Pre-flight validator

FFmpeg

Deterministic engine

1,200 free

Credits to start

FAQ

Frequently asked questions

Does Zvid generate video from a text prompt? +

No. Zvid does not do text-to-video, diffusion, or generated pixels. An AI agent authors a structured JSON project — typically through the MCP server — and Zvid renders that JSON deterministically with FFmpeg.

How does an AI agent connect to Zvid? +

Through the MCP server, which exposes 30 tools with schema-aware authoring to agents like Claude Code, Cursor, and Codex. A free pre-flight validator lets the agent confirm a project is valid before any credits are spent.

Why does deterministic rendering matter for AI output? +

Because the engine is FFmpeg rather than a headless browser, the same JSON always renders the same frames. An agent can regenerate a document and trust the video matches, which matters when a model produces content at scale for your brand.

Explore more: see every use cases

Ready to ship video at scale?

Start free — your first render is only a JSON payload away.