Generate 3D videos from your AI agent with @olo-studio/studio-mcp.

By the Olo Studio team · Updated November 14, 2026

Short answer

Install the local-first @olo-studio/studio-mcp server and universal skill with npx skills add faustodariopenamondragon-web/OlomniAI --skill olo-studio. Your agent discovers 7 MCP tools, compiles deterministic SceneSpec v2 contracts with 10 cinema cameras and 15 Post-FX, self-inspects keyframes via previewFrames, and exports MP4, WebM Alpha, or Chroma MP4 locally at $0 cloud GPU.

Step by step

Connect the MCP server and Skill in one command. In Claude Code run claude mcp add olo-studio -- npx -y @olo-studio/studio-mcp && npx skills add faustodariopenamondragon-web/OlomniAI --skill olo-studio (for Codex use codex mcp add; in Cursor or Antigravity paste the mcpServers JSON block).
Optional: set your olo_pat_... token. The Free local sandbox requires no key; to unlock zero watermark, ElevenLabs voices, or batch exports, create a Personal Access Token at olomni.com/studio?auth=mcp and set OLO_STUDIO_API_KEY.
Inspect capabilities via olo_describe_capabilities. Your agent reads the live catalog: 10 camera presets, 15 Post-FX, 5 LUTs, 10 transitions, 16 3D gestures, 14 HUD cards, and 4 device mockups.
Compile and render locally. Ask for a 9:16 or 16:9 launch reel: the agent calls olo_compile_scene (olomni.studio.scene.v2) and olo_render_video with previewFrames: true to verify framing before handing you the .mp4 and .srt.

What the 7 tools in @olo-studio/studio-mcp v1.3.1 cover

olo_describe_capabilities: live SceneSpec v2 capability manifest and starter recipes.
olo_get_entitlements: checks your tier (Free, Creator & Builder Pro, or Brand & Agency) and voice quotas.
olo_generate_script: turns URLs, chats, or changelogs into annotated Olo: / Yo: beats.
olo_compile_scene: compiles deterministic olomni.studio.scene.v2 (and v1-compatible) JSON without launching a browser.
olo_render_video: renders in local headless Chromium (WebGL2 + WebCodecs) at up to 60 fps with previewFrames: true and .srt/.vtt captions.
olo_render_batch and olo_voice_clone_list: 50-ad CSV batches and voice clone management.

Why local-first and deterministic matters

All 3D rendering runs on your machine at $0 cloud GPU cost and produces a reproducible sceneHash in 9:16 (1080×1920), 16:9, 1:1, or 4:5. Explore the full interactive reference at olomni.com/en/mcp.

Frequently asked questions

Do I need a paid plan to try @olo-studio/studio-mcp?

No. The local sandbox and SceneSpec v2 compiler work for free with no account or credit card (up to 30 s in 9:16 with a discrete watermark). Creator & Builder Pro ($19/mo) and Brand & Agency ($149/mo) unlock via OLO_STUDIO_API_KEY (olo_pat_...).

Which MCP clients are supported?

It works over stdio with Claude Code, Antigravity, OpenAI Codex, Cursor, Windsurf, Claude Desktop, and direct CLI execution via npx -y @olo-studio/studio-mcp.

What does previewFrames: true do?

It saves up to 6 JPEG keyframes alongside the video so your AI agent can visually inspect framing, HUD cards, and subtitles before delivering the final MP4.

Keep reading

How to make a Product Demo video with transparent WebM Alpha & 3D device mockups (Demo PiP)Overlay Olo with a transparent background (WebM Alpha or Chroma MP4) on top of SaaS screen recordings, or render inside iphone_16, macbook_pro, and browser_window mockups at 60 fps. How to render up to 50 A/B video ads for Meta & TikTok from Bulk CSV or MCP (.ZIP)Build 3×3 creative matrices and up to 50 vertical 9:16 ad variations combining hooks, voices, sets, and CTAs using Bulk CSV in Studio or olo_render_batch at $0 cloud GPU. How to upload your custom 3D brand mascot (.GLB / .GLTF) with auto-rigging and 7 socketsImport a custom .GLB or .GLTF character into Olo Studio, validate it with inspect-glb, attach props to 7 skeleton sockets, and drive all 21 states and 16 gestures instantly.