Scenetra MCP Server

Connect once.
Create anywhere.

The Scenetra MCP server turns Claude Code, Cursor, and Claude Desktop into image and video studios. Your agent browses the models, designs the workflow, builds the board, and hands you finished renders — on your account, in your studio.

claude mcp add scenetra --transport http https://app.scenetra.com/api/mcp --header "Authorization: Bearer sct_YOUR_TOKEN"

Get your token in Settings → MCP / API Access

Get your token How it works →

Works with Claude Code · Cursor · Claude Desktop · any MCP client

Every frontier model maker, one canvas

Three steps

From zero to first render.

Create a token

One click in Settings → MCP / API Access. It's shown once, stored hashed, and revocable anytime — the only credential you need.

Paste one command

Pick your tool below; setup is a single paste. No OAuth dance, no config surgery.

Ask for what you want

“Build me a three-style product shoot.” Your agent plans the board, wires the nodes, runs it, and reports back with links.

claude mcp add scenetra --transport http https://app.scenetra.com/api/mcp --header "Authorization: Bearer sct_YOUR_TOKEN"

Paste in your terminal

In practice

One request. A finished board.

This two-mood lighthouse study was built end-to-end by Claude Code over MCP — prompts, concatenators, generators, groups — then executed and polled to completion in twelve seconds. Nobody touched the canvas.

7 nodes wired2 branches in parallel12s run

Build a board that paints the same lighthouse in two moods — dusk and storm — and run it.

→ get_guidelines
→ get_model_schema z-image-turbo
→ execute_board “Lighthouse Moods” · RUNNING
→ get_run_status COMPLETED · 2 images

Done — two lanes share one style node; dusk and storm rendered in parallel. Board and images below.

Lighthouse at dusk, oil painting — generated by an agent through the Scenetra MCP — Dusk — rendered by agent

Lighthouse in a storm, oil painting — generated by an agent through the Scenetra MCP — Storm — rendered by agent

Open the board it built →

What agents can do

Seven tools.
The whole studio.

Knows the house rules

Agents read Scenetra's board-building guidelines before they build — lanes, shared prompts, wiring patterns, common pitfalls.

Browses 50+ models

The full catalog with live pricing — image, video, and audio, from fast drafts to cinematic renders.

Reads every schema

Exact input fields, handles, and costs per model, generated from the same registry the app runs on. Never out of date.

Drafts blueprints

Boards without execution — templates and drafts that cost nothing until you choose to run them.

Runs and reports

Kicks off generation, polls progress while branches run in parallel, and returns image and video URLs when the run completes.

Extends live boards

Adds new stages to boards you already have, consuming outputs already on the canvas — append-only, never destructive.

Bills like you'd expect

Runs use your credits or your own provider keys — same rates as the app, no separate plan.

Browse the model catalog →

Questions

The fine print, minus the fine print.

There's no separate MCP plan. Runs started by your agent bill exactly like runs in the app — subscription credits, or your own provider API keys on the lifetime license.

Tokens are shown once and stored hashed. Revoke any token instantly in Settings; connected tools lose access on their next request.

Anything that speaks MCP over HTTP. Claude Code and Cursor connect natively; Claude Desktop connects through the mcp-remote bridge. One token works everywhere.

No — that's the point. You talk to your agent; the agent reads the guidelines and schemas and writes the graph. The board lands in your studio, editable like any other.

All of them — the same 50+ image, video, and audio models as the app, with live per-run pricing.

Bring your agent.
Leave with renders.

Get your token Browse models