OriginShorts MCP Server

Connecting

Two ways in. Marketplace clients use OAuth and need nothing from you but a login; everything else can carry an API key directly.

With OAuth

In ChatGPT, Claude or any client that supports remote MCP servers, add https://originshorts.com/mcp as a connector and approve the consent screen. The server registers clients dynamically, so there is no key to paste and nothing to configure.

Discovery and endpoints, if your client asks for them explicitly:

With an API key

Create a key on your account page and send it as a bearer token. Keys start with os_live_ and carry the permissions of the account that made them.

Authorization: Bearer os_live_YOUR_KEY

Treat a key like a password: it can spend your credits. Keep it out of shared configs and source control.

Client configuration

Copy the block for your client. Replace os_live_YOUR_KEY with your own key, or drop the header entirely if you are connecting over OAuth.

Claude Code

claude mcp add originshorts --transport http https://originshorts.com/mcp \
  --header "Authorization: Bearer os_live_YOUR_KEY"

Claude (web and desktop app)

Customize → Connectors → + Add → Add custom connector. Name it OriginShorts, paste https://originshorts.com/mcp, and sign in when asked (OAuth). No key required. On Team and Enterprise plans, an owner adds it first under Organization settings → Connectors.

Cursor

In ~/.cursor/mcp.json, or .cursor/mcp.json in a project:

{
  "mcpServers": {
    "originshorts": {
      "url": "https://originshorts.com/mcp",
      "headers": {
        "Authorization": "Bearer os_live_YOUR_KEY"
      }
    }
  }
}

ChatGPT

Turn on Developer mode (Settings → Apps & Connectors → Advanced settings), then create a connector pointing at https://originshorts.com/mcp with OAuth and approve the sign-in screen. No key required. Needs a paid ChatGPT plan; on Business and Enterprise workspaces, an admin enables custom connectors first.

Workflow

Generation is asynchronous. One call starts it, then you poll. A video usually takes one to three minutes, a few more when animated.

1. Create

   Call create_video with a topic (we write the script) or a script of your own. It returns a video id straight away.

2. Wait

   Call get_video_status. It waits on our side for up to about 40 seconds and returns the moment the video is done, so an assistant just calls it again until it is. No pauses between calls. Status moves queued → processing → rendering → completed, with a plain-language stage and percent along the way.

3. Download

   On completion the response carries a download_url for the MP4. It is valid for one hour. Poll again for a fresh link.

Tools

Twenty-eight tools: six to make and track videos, seventeen to look at and edit them, five to post them. Every response is JSON except view_video, which returns the image itself; errors come back as readable text with the reason and what to do about it.

Editing

An assistant edits a video the way a person does in the editor, pays the same credits, and renders once at the end. Edits never re-render on their own: a finished video that has been edited reports unrendered_changes: true until render_video. Slow edits and renders queue and run in order, and get_video_status waits for them. Images, animations, voiceovers and portraits an edit replaces stay in the video's History in the OriginShorts app, where you can switch back.

Posting

An assistant can post a finished video to the user's YouTube, TikTok and Instagram, now or at a set time, as the app does: public unless it sets the YouTube or TikTok privacy (Instagram posts are always public), with the titles and captions written for the video unless it writes its own. Connecting an account takes a link the user opens and approves; they don't need to sign in to OriginShorts.

Allowed values

The authoritative list is whatever get_video_options returns. These are the fixed enums.

Credits

Plans are on the pricing page.

Errors

Failures return readable text rather than a bare code, so an agent can usually recover without help.