Skip to main content
Two ways to work with Preview from outside the app: the CLI, a JSON-first command-line client for scripting and coding agents, and the Preview MCP, which lets AI apps like ChatGPT, Claude, and Cursor browse, generate, and build in your projects directly. Both run as you, with the same project access and credits as the app.

​
Install the CLI

Or through npm:
Either way, the CLI keeps itself up to date: on startup it checks for a newer version, updates through whichever channel you installed with, and runs your command as usual. Opt out with --no-update.

​
Log in

Run preview auth login. It opens a browser to confirm a device code, then stores your credentials locally; every other command works once you’re signed in.

​
What the CLI is useful for

  • Scripting your production data: list, create, and update projects, stories, scenes, shots, and takes from the command line or a script, without opening the app.
  • Uploading and managing media: attach files to shots, manage library items (Characters, Locations, Props), and organize project assets.
  • Connecting coding agents: give Claude Code, Codex, or similar tools direct, structured access to a project so they can help plan and build it out.
  • Serving the Preview MCP locally: preview mcp init wires Claude Desktop, Claude Code, Cursor, and VS Code up to your projects, covered below.
  • Batch operations: run a list of operations from a single file instead of one command at a time.
Every command supports --json for scripting; destructive commands require an extra --yes confirmation flag.

​
Connect ChatGPT or claude.ai

No install needed. Add a connector with this URL:
Connecting opens a browser sign-in to Preview; approve it and you’re done.

​
Connect Claude Desktop, Claude Code, Cursor, or VS Code

These run the MCP locally through the CLI. With the CLI installed and signed in:
That configures all four clients in one go. Use --target claude-desktop, --target claude-code, --target cursor, or --target vscode for just one, or --print to see the config snippet without writing anything.

​
What the MCP can do

  • Browse and search projects: project snapshots, canvases and canvas groups, takes and their comments, and context for any sequence, shot, or library item.
  • Generate: images and video on any model, singly or in batches, with cost estimates before running, onto the canvas or into an existing group.
  • Build structure: create and update sequences and shots, assign takes to shots, and set statuses.
  • Manage the library: create Characters, Locations, and Props, attach images, and link them to shots.
  • Write Scripts: list, create, import (PDF, Final Draft, Fountain, or text), and edit a project’s Scripts.
Files uploaded through the CLI or MCP are saved to the project’s asset library, so they show up in the app’s reference pickers like any other upload. In ChatGPT and Claude, a generation renders as an interactive card in the conversation: a hero image or video, a strip of every take, and live progress while it runs. Click a thumbnail to make it the hero and select it in Preview.
Generating through MCP spends credits exactly like generating in the app, and every result lands in your project’s Files view and History like any other take.