Documentation
Idea to Motion is a Model Context Protocol server that turns an AI client you already use into a video studio. You connect one endpoint, ask for a video in chat, and your assistant does the creative work — it writes the script, plans every scene and writes the animation code. The platform synthesizes the voiceover, aligns it to word-level timestamps, renders the composition and returns an MP4.
No server-side model makes a creative decision. That is the architectural difference from a text-to-video API: the thinking happens in the client’s own model, on the user’s own context, and voice synthesis and transcription run on the user’s own provider key. Creating and previewing are therefore free and unlimited, and the only thing the platform charges for is rendering a finished file.
One Streamable HTTP endpoint serves every client. It implements the 2026-07-28 MCP specification and serves 2025-era clients on the same URL, so there is no separate legacy address to configure. Serving is stateless and per-request: each POST builds a fresh server instance bound to the authenticated user, which is what lets several agents work on the same video in parallel without crossing wires.
https://api.ideatomotion.com/mcpCopy-paste config for each client — including one-click deeplinks for Cursor and VS Code — is on the connect page.
There are two ways to authenticate, and which one you use depends entirely on whether your MCP client can send a custom header. Both resolve to the same account and the same set of jobs, so a video created from a terminal client is editable later from a desktop one. An unauthenticated request is answered with a 401 carrying aWWW-Authenticateheader that points at the OAuth metadata, so a client can discover the flow on its own.
vai_A long-lived key you mint on the platform and paste into your client's config as an Authorization header. This is the mode for clients that can send headers — Claude Code, Cursor, VS Code, Codex, Windsurf and the mcp-remote bridge. Keys are stored as a sha256 hash, so a key is shown once at creation and never again.
Authorization: Bearer vai_…vao_A full OAuth 2.1 authorization-code flow with PKCE and dynamic client registration, for clients that accept only a URL and no headers — Claude's web and desktop custom connectors. Discovery metadata is published under /.well-known, access tokens live one hour, and refreshing rotates the pair.
Authorization: Bearer vao_…The full workflow is published as an open, host-agnostic SKILL.md that any agent can read — Claude Code, Cursor, Codex, Windsurf, Gemini CLI and VS Code all load skills from the same place. It documents the whole loop end to end: which brief to fetch at each stage, how to direct scenes so consecutive ones differ, how to write and verify a scene, and how to recover from a validation failure. It is plain text, served over HTTP, and reading it does not require an account.
The same knowledge also rides in the server’s instructions field at initialize, so a client with no skill support still gets the workflow with zero install.
Every job is one of three kinds, chosen at creation. Aspect ratio and duration are independent of the kind and of each other, so a fifteen-second widescreen clip and a three-minute vertical explainer are both reachable. A genre style — a product launch, a news article, a systems explainer — can be layered on top, and it rewrites every brief the server hands back rather than merely changing colours.
The server exposes 31 tools across 8 phases, from account setup through to the adjustments you make on a video that is already finished. The front half of the pipeline is deliberately data-driven — one set of plan tools serves a script, a shot list and a cut plan alike — so supporting a new kind of video adds no tools at all and an agent never has to learn a new vocabulary.
An agent can create a video, plan it, write every scene, preview it, re-frame it and remix its audio — but it cannot render the final MP4. Exporting is the only step that spends real money, and the agent cannot see the user’s balance before it acts, so an export tool would mean spending on a quality the user never picked at a moment they never chose. The user exports from the platform instead. Re-exporting a video they have already paid for is free, so pointing them back to the browser costs them nothing.