Engineering & product

The JSON your agent writes, rendered frame-accurate.

Zvid gives your agent 30 MCP tools, schema-aware authoring, and a free validator to write JSON that renders to video and images — deterministically, with FFmpeg, not text-to-video.

Agent-native

Give your agent a video backend it can trust

Zvid is not text-to-video. There are no diffusion models and no generated pixels — your agent authors a JSON project and Zvid renders it with FFmpeg, so the same document produces the same frames every time. The MCP server exposes 30 tools to clients like Claude Code, Claude Desktop, Codex, and Cursor, with schema-aware authoring that keeps the model on valid fields and elements.

Because the pipeline is deterministic, an agent can reason about the output, diff it, and rely on it. Run any project through the free pre-flight validator first: it resolves your templates, applies your plan limits, and returns errors by field without spending a credit — then a repair tool helps fix what it flags. When the JSON is clean, one authenticated POST returns a hosted mp4 or image at a cdn.zvid.io URL.

What you get

Built for agent workflows

30 MCP tools

Point Claude Code, Claude Desktop, Codex, or Cursor at the MCP server and your agent gets 30 tools to author, validate, and render Zvid projects.

Validate before you spend

The free pre-flight validator resolves templates and applies your plan limits without spending a credit, so invalid JSON never reaches a billed render.

Deterministic renders

FFmpeg renders one JSON document to the same frames every time — reproducible, frame-accurate output an agent can diff and trust, not generated pixels.

Schema-aware plus repair

Schema-aware authoring keeps the model on valid fields and elements, and a repair tool fixes what the validator flags before you retry.

Retries stay safe

Send an idempotency key with each POST so an agent can retry on timeout without risking a duplicate job or double-charged credits.

Batch and callbacks

Queue up to 500 renders in one call with best-effort per-item validation, then key on HMAC-SHA256-signed render.completed and render.failed webhooks.

30

MCP tools

FFmpeg

Deterministic engine

Free

Pre-flight validator

500

Renders per call

FAQ

Frequently asked questions

Does Zvid generate video from a text prompt? +

No. There is no diffusion model and no text-to-video — your agent authors a JSON project and Zvid renders it deterministically with FFmpeg. 'AI-generated' here means an LLM writes the JSON, and the same JSON always produces the same frames.

How does an agent connect to Zvid? +

Through the MCP server, which exposes 30 tools to clients like Claude Code, Claude Desktop, Codex, and Cursor. There are also n8n community nodes, a Make app, and Python and TypeScript SDKs — see the docs for setup.

Can I check a project is valid before spending credits? +

Yes. The free pre-flight validator resolves your templates and applies your plan limits without billing, returning errors by field, and a repair tool helps fix them. The free plan also includes 1,200 credits with no card required.

Explore more: see every roles

Ready to ship video at scale?

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