Videos from your codebase: Ozor in Claude Code, Cursor and Codex

Your coding agent knows exactly what shipped this sprint. It read every diff, wrote half of them, and can tell you which change customers will notice. It is also, it turns out, a capable motion designer, because an Ozor video is code: a React component per scene, rendered frame by frame. Connect Ozor to Claude Code, Codex, Cursor or VS Code and the release video, the PR demo and the architecture walkthrough come out of the same session as the code.

Why coding agents make good video makers

An Ozor scene is a React component with a single clock: the current frame. The agent writes it with inline styles and a shared library that includes Framer Motion, GSAP with its plugins, Recharts, D3 and Three.js. That is a coding agent's home turf, and the scene guide it reads first spells out the rest: no wall-clock APIs, no CSS transitions, motion that keeps moving, type that stays readable.

The bigger advantage is context. A marketing tool has to be told what changed; your coding agent already knows, down to the copy on the new button. And because Ozor keeps every scene separate, fixing the video works like fixing code: change the one scene that is wrong, leave the rest alone.

Ozor's scene library helps here. There are 121 tested example scenes the agent can search by topic and adapt, including a terminal reveal, a deploy pipeline, a branch merge, a file tree scaffolding itself, an API request and response, a dependency graph, code turning into UI, and an uptime wall.

Setup for each agent

All four use the same remote server and the same OAuth approval page in Ozor. You will find one-click versions of each in Ozor under Developer → Connect AI assistants.

claude mcp add --transport http ozor 'https://api.ozor.ai/mcp'

codex mcp add ozor --url 'https://api.ozor.ai/mcp' && codex mcp login ozor

{
  "mcpServers": {
    "ozor": { "url": "https://api.ozor.ai/mcp" }
  }
}

{
  "servers": {
    "ozor": { "type": "http", "url": "https://api.ozor.ai/mcp" }
  }
}

To check it worked, ask the agent: "Which Ozor tools do you have?" It should list tools like get_scene_guide, start_video, add_scene, preview_scene and export_video.

Four workflows worth stealing

A habit that pays off: ask the agent to call preview_scene on anything UI-heavy. It renders real frames and returns them as images, so the agent can see that the recreated screen looks right before moving on, the same way it would run the tests.

Running it unattended

OAuth is the right choice on your own machine. For a pipeline, a scheduled job or a headless agent, create an Ozor API key under Developer → API Keys and send it as a bearer token. The server accepts either; both act as your account.

claude mcp add --transport http ozor https://api.ozor.ai/mcp \
  --header "Authorization: Bearer $OZOR_API_KEY"

Keep the key in your CI secrets, create one per pipeline so each can be revoked on its own, and have the job ask for an export at the end so it gets a watch link to post. If you would rather not have an agent in the loop at all, the n8n node and the REST API drive the same engine.

Frequently asked questions