Connect ChatGPT, Claude, or Cursor to an AI Video Studio (MCP Guide)

You already pay for ChatGPT Plus, Claude Pro, or Cursor. That subscription is very good at writing, planning, and reasoning — and completely unable to render a shot of your character walking through rain at dusk.

Meanwhile, most AI video tools ship their own built-in assistant, charge you again for it, and cap how many messages you get.

There is a better arrangement. Connect the AI you already have to the studio that does the rendering. The assistant does the thinking on your existing subscription; the studio does the generating. This guide shows you how, using Hyperdream's remote MCP server.

What this actually gets you

Once connected, you can say this in ChatGPT or Claude:

"Make me a 30-second ad for a cold brew coffee brand. Moody, cinematic, one barista character. Write the script, design her, storyboard it, and render the shots."

and the assistant will work through the real production pipeline — writing the script, locking a visual style, casting and generating the character's reference sheet, building scenes and shots, then rendering video. Everything lands in your Hyperdream projects, where you can open it, edit it, and export it like any other project.

The division of labour is the point:

Who does what	Cost
Planning, writing, direction, iteration — your assistant	Your existing AI subscription
Image and video generation — Hyperdream	Credits, only when something is actually rendered

Because the reasoning no longer runs on Hyperdream's account, there are no assistant message limits on this path. You can plan a feature-length treatment across a hundred messages and spend nothing until you ask for a render.

What is MCP?

MCP — the Model Context Protocol — is an open standard for letting AI assistants use external tools. A remote MCP server is a web address that exposes an app's capabilities as tools an assistant can call.

It matters here because it's shared. One connection works across ChatGPT, Claude, Cursor, Antigravity, Windsurf, Cline, and anything else that speaks the protocol. You are not learning a vendor-specific plugin system.

Connecting ChatGPT or Claude

These run a proper sign-in flow, so all you need is the URL.

1. In Hyperdream, open Settings → Connect your AI tools and copy the server URL (https://hyperdream.co/api/mcp).
2. In Claude: Settings → Connectors → Add custom connector. In ChatGPT: Settings → Connectors → Add.
3. Paste the URL. You'll be sent to Hyperdream to sign in and approve the connection.
4. Approve it. The assistant now has the tools.

Your password is never shared with the assistant, and you can disconnect any app from the same Settings page.

Connecting Cursor, Claude Code, or Antigravity

Developer tools authenticate with a header instead, so they need a key.

1. In Settings → Connect your AI tools, switch to the second tab and create a key. Copy it immediately — it is shown once and never again.
2. Add the server to your tool's MCP configuration:

{
  "mcpServers": {
    "hyperdream": {
      "url": "https://hyperdream.co/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_KEY_HERE"
      }
    }
  }
}

For Claude Code, one command does it:

claude mcp add --transport http hyperdream https://hyperdream.co/api/mcp \
  --header "Authorization: Bearer YOUR_KEY_HERE"

The workflow that actually works

The pipeline is ordered, and skipping steps produces worse film — the same as in real production.

1. Script first. Give the assistant the brief and let it write the script into a new project.
2. Style before anything visual. The style pass locks the aesthetic and aspect ratio for the whole film. Everything generated afterwards inherits it.
3. Cast, then generate every character image. This is the step people skip and regret. A character with no reference sheet has no identity to lock to, so they come out looking like a different person in every shot. Hyperdream will refuse to render a shot whose characters have no images — that refusal is protecting your credits.
4. Environments next. Generating the environment plate for a scene keeps every shot set there visually continuous.
5. Storyboard, then shots. Scenes first, then the shots inside them.
6. Render. Video takes one to three minutes per shot, so renders run in the background — the assistant starts a job and checks back until it's done. Ask for one shot first, look at it, then commit to the rest.

Tips that save credits

• Preview a frame before animating it. Generating a still frame is far cheaper than a video. Approve the frame, then animate exactly that frame.
• Ask before you commit in bulk. "How many credits will rendering all eight shots cost?" is a question the assistant can answer before you spend.
• Fix the character, not the shot. If someone looks wrong across several shots, regenerate their reference sheet rather than re-rolling each shot.
• Failed renders refund automatically. If a generation fails, the credits come back — you do not need to chase it.

Does this replace the built-in assistant?

No, and it isn't meant to. Hyperdream's own assistant is still there and still the fastest way to start if you don't have an AI subscription or don't want to configure anything.

Connecting your own AI is for the people who hit the ceiling: heavy users, anyone already living in Cursor or Claude all day, and anyone who would rather spend their credits on pixels than on chat.

Connect your AI tools in Hyperdream →