Connect Your AI Assistant

What this gives you

Once connected, your assistant can do everything up to and including starting a production, and it can read a finished plan back to you so you can approve it without leaving the conversation. Specifically it can:

• read the design vocabulary the platform accepts — styles, tones, audiences, voices, languages, target platforms;
• dry-run a batch of briefs and show you what would be built, without creating anything;
• create drafts, which is free;
• quote a production's price;
• start it, against that quote — the only thing that spends credits;
• show you the storyboard, and approve it once you say so;
• poll progress and hand you a download link at the end.

It cannot publish anything anywhere, and it cannot save a reusable brand preset. Both of those stay in the app on purpose.

Before you start

• A StudioCut.Video account with credits — either plan credits from a subscription or purchased credits.
• An MCP client. Claude Desktop and Claude Code both work; so does any agent framework where you can set a URL and an Authorization header.
• About five minutes.

Step 1 — Enable API access

The first time you open MCP and API Keys, the page asks you to make one decision before it will issue anything, under Read this before you continue:

• Choose what happens to delivered output. Either Yes — delete it after N days, or No — keep everything. There is no default and no silent choice, because both defaults are wrong: one grows your storage without telling you, the other deletes your videos without asking. Pick either — you can change it whenever you like.

Then press Enable API access. See automatic deletion below for exactly what that setting does and when the clock starts.

Step 2 — Generate a key

Press Generate key and fill in three things:

• Key label — a name for your own benefit. Use the client and machine, e.g. "Claude Desktop — laptop". When you have four keys you will be glad you did.
• Expires after — in days, up to a maximum of 30. Shorter is safer; rotation (below) means short-lived keys cost you nothing in downtime.
• Acts on — My personal productions, or one specific agency. This is the single most important field on the form: it decides everything the key can see and spend, it cannot be changed afterwards, and there is no way for the assistant to switch context at call time. If you need both, generate two keys.

Step 3 — Paste it into your assistant

Add StudioCut.Video to your client's MCP server configuration:

{
  "mcpServers": {
    "studiocut": {
      "url": "https://app.studiocut.video/mcp",
      "headers": { "Authorization": "Bearer YOUR-KEY-HERE" }
    }
  }
}

One HTTPS endpoint, one header. There is no separate account id or workspace id to supply — the key already knows which account or agency it acts on.

Step 4 — Check the connection

Restart your client and ask it something read-only, for example:

"What video styles and languages does StudioCut support?"

A working connection answers from the live catalogue. If instead your client reports an authentication failure, jump to common problems — the fix is almost always the key itself.

Your first video, start to finish

A normal first run looks like this. You are in the conversation for two of these steps, and only two.

1. Describe what you want. "Make a 60-second vertical video explaining our refund policy, friendly tone, for existing customers."
2. Ask for a dry run first. The assistant validates the brief and shows you what would be built — normalised the way the platform will actually read it. Nothing is created and nothing is charged.
3. Let it create the draft. Still free. A draft can sit for a week if you change your mind.
4. Look at the price. The assistant quotes the production and tells you the figure. This is your first decision. The quote is live for five minutes.
5. Say start. Credits are deducted once, for the amount you were shown.
6. Wait for the storyboard. The assistant polls and tells you when the plan is ready, then reads it back — scenes, script, timing.
7. Approve it, or fix it first. This is your second decision, and the point of no return. To change something, edit the production in the app, then tell the assistant to approve. Phases 2 to 5 then run unattended.
8. Collect the link. When it is done the assistant fetches a download link for you to open.

Where your assistant's videos appear

They get their own list on the MCP and API Keys page, under API productions — not the main production grid and not the Media Library. That separation is deliberate: an overnight batch of twenty should not bury the three videos you are working on by hand.

Each row shows the reference, its status, whether it has been delivered, its size, and a preview thumbnail. You can select rows and use Delete selected to remove them permanently, which frees the storage immediately.

Everything else is unchanged: the videos are yours, they download the same way, and they cost the same to store.

What it costs

• Which credits. Assistant work spends your plan credits and your purchased credits.
• When you are charged. Once, when the production starts — exactly as in the app. Drafting, validating, quoting, polling, reading the storyboard and fetching a download link are all free.
• What your plan does and doesn't control. Plan feature gates don't apply to assistant-driven production, and this work doesn't consume the monthly video allowance you bought for the app. What still applies to everyone: your storage limit, the content policy, up to five extra languages per production, and the rate and concurrency limits on the key.
• The channel premium. A production started through an assistant carries a 12.5% premium, and only when both of these are true: the quality tier is 2 or higher, and the target duration is over 30 seconds. Shorter or lower-tier jobs are priced exactly as in the app. The premium is shown as its own line in the quote breakdown before anything is charged — it is never a surprise on the invoice.

When the balance runs low

You can set a threshold on the API page. When your spendable balance falls below it, StudioCut.Video prepares a top-up invoice and emails you the payment link; the credits arrive when you pay it. No card is ever charged automatically — there is no saved payment method on this platform, and even subscription renewals are invoice-then-pay.

There is also a cap on how many of those automatic invoices can be raised in a month, so a runaway integration produces one email and a halt rather than a stack of invoices.

Automatic deletion of delivered output

This is the setting you chose in Step 1. If you turned it on:

• The clock starts when a video download link is issued — that is, when you or your assistant actually take delivery. Nothing else starts it. Loading a thumbnail does not.
• Output you never downloaded is never automatically deleted. There is no backstop sweep.
• Deletion removes the files, not the record. The production stays in your list showing as expired, and its name, cost and dates remain. Asking your assistant for a download link on an expired production gets a clear "the files were purged" answer rather than a silent failure.
• You get a warning email partway through the window, and the API productions list shows a countdown per row.
• Any production can be marked Keep to exempt it permanently, and you can change the window or switch the whole thing off at any time.

If a video might matter later, download it or mark it Keep. That is the whole rule.

Looking after your key

• Keys expire — 30 days at the outside. Treat expiry as a scheduled event, not an incident. Reminder emails arrive a week and a day beforehand.
• Rotate rather than replace. Rotate issues a new key immediately and keeps the old one working for a grace period (24 hours by default), so you can update a config file on your own schedule instead of during an outage. The replacement carries the same binding, the same ceiling and the same history.
• Unused keys are withdrawn. A key nobody has used for a fortnight is revoked, after a warning email. Any authenticated request resets that clock, so a live integration is never affected — this only catches credentials that have been forgotten.
• Revoke anything you cannot account for. Revocation takes effect immediately, everywhere.
• One key per client. Sharing one key across your laptop, your server and a colleague makes the audit log useless and turns one compromise into three.

Common problems

What you see	What it means	What to do
Your client reports an authentication failure	The key is unknown, expired, the wrong kind, or its account was archived. These are deliberately indistinguishable	Generate a fresh key on the MCP and API Keys page, and check it did not come from the Odoo backend
The assistant says a tool doesn't exist	It cached the tool list from an earlier session	Start a new session so it re-reads the list
"Rate limit" while it is waiting for a video	It is polling too tightly	Tell it to check every 30 seconds or so; the error says how long to wait
The production seems stuck	It is waiting for the storyboard approval — the one human gate	Ask for the storyboard, then approve it
"A valid confirm token is required"	The quote is older than five minutes, or was issued for something else	Ask for a fresh quote, agree the price again, then start
"The cost of this production changed"	The price moved between the quote and the start	Look at the new figure and decide again — this is the guard working
A second video won't start	One production per key at a time, and a storyboard awaiting your approval holds the slot	Approve or finish the first one

The full error catalogue, including the machine-readable codes, is in the tool reference.

See also

• MCP Tool Reference — every tool, argument, error code and limit, for building your own client.
• AI Assistants (MCP) — what the integration is and why it is shaped this way.
• Storyboard Review — what you are actually approving, and how to change it.
• The Create Wizard — the same fields your assistant fills in, with every option explained.
• Five things people build with it — worked examples end to end.