The brief already lives in the chat
You researched the topic with your assistant, argued about the angle, settled on a title. Copying all of that into a form is the least interesting part of the job — and the part where the detail gets lost.
Integrations · MCP server
Make videos from your AI assistant.
StudioCut.Video speaks MCP — the protocol assistants use to reach real tools. Point Claude, Claude Code or any MCP client at your account and it can read the design vocabulary, draft a batch of productions, price them, start one, and hand you the storyboard to approve. Drafting is free. Nothing renders, and nothing is charged, until you agree to a quoted price. It's part of the StudioCut.Video capability catalog.
Why it matters
Your assistant already knows what the video should be about. Now it can make it.
Batches are where the time goes
One video is a five-minute form. Twenty videos is an afternoon of the same five minutes. An assistant fills twenty briefs while you read the dry run.
Automation usually means losing control
The reason people don't automate video is the fear of waking up to forty renders they didn't want. That fear is what the price quote and the approval gate are built to remove.
What it looks like
A production, in one conversation
No new interface to learn. You talk to the assistant you already use, and it calls StudioCut.Video on your behalf.
"Turn last week's three blog posts into 60-second vertical videos, same brand as usual."
The assistant reads the accepted styles, tones, voices and languages from the platform first, so it briefs in vocabulary StudioCut.Video actually understands.
"Here's the dry run — three productions, portrait, your brand kit. Shall I create them?"
The dry run creates nothing and costs nothing. You see exactly what would be built, normalised the way the platform will actually interpret it.
"Drafts created. The first one quotes at X credits. Start it?"
Drafts are free and can sit indefinitely. The quote is a real figure with a five-minute life, not an estimate reconciled later.
"Go. Show me the storyboard when it's ready."
The assistant polls, then reads the finished plan back to you — scene by scene — so you can approve it in the chat instead of opening a browser.
"Scene 4 is wrong. Fix it in the app — then approve."
Approval is the point of no return, and it is always yours. Phases 2 to 5 then run unattended.
"Rendered. Here's the download link."
The link is yours to open. The other two drafts are still sitting there, unstarted and uncharged, until you want them.
The toolbox
Twelve tools. Exactly one of them spends money.
Everything an assistant can do here is on this list. There is no general-purpose escape hatch, no raw database access, and no tool that was added because it was easy.
list_design_options
The design vocabulary the platform accepts — styles, tones, audiences, voices, languages, platforms. Read live, so it is never out of date.
list_productions
What is in the account, filterable by state and style.
get_production
One production's settings in full.
get_production_status
Where it has got to, and what the next action is. This is the tool an assistant polls while a video is being made.
get_storyboard
The finished plan — scenes, script, timing — so the one human decision can be made on evidence, in the conversation.
get_balance
What this key can actually spend, its ceiling and the spend so far. Called before a batch is designed, "not enough credits" arrives while it is still useful.
get_download_url
A link to the finished video, for you to open.
validate_productions
Checks a batch of briefs and reports exactly what would be built — and what it would fix or reject. Creates nothing.
create_productions
Turns briefs into drafts, up to twenty in a call. Drafts cost nothing and can sit as long as you like.
estimate_cost
The price for one production, broken down, plus the token that authorises starting it. The token lives five minutes.
start_production
The only tool that spends. It refuses without a live quote token, and refuses again if the price moved since the quote — returning the new figure instead of charging it.
approve_phase1
Approves the storyboard. After this, phases 2 to 5 run unattended to a finished video.
The safety story
An agent cannot take a brief all the way to a published video
Three limits do the work, and none of them are settings you have to remember to switch on.
One human approval, always
The pipeline stops at the storyboard and waits. An assistant can read that plan to you and argue for it, but it cannot approve it on your behalf — and the production holds its slot while it waits, so nothing else starts behind an unreviewed plan.
A price, then a confirmation
Starting requires a token issued by a quote in the last five minutes. If the price changed in between, the start is refused and the new number is reported. There is no path where you are charged a figure you were never shown.
Nothing goes out to the world
There is no publish tool. Nothing pushes to YouTube or any other platform, because that is irreversible and outward-facing — it stays in the app, where a person does it on purpose.
Money and limits
Your plan doesn't gate it. Your ceiling does.
Assistant-driven production spends your plan credits and your purchased credits, and it doesn't eat the monthly app allowance you bought separately.
Feature gates don't apply here
Work started through an assistant isn't held back by your plan's feature tiers, and it doesn't count against the monthly video allowance you use in the app. What you can afford is what you can make.
Set a ceiling per key
Each key can carry a credit ceiling for a day, a week or a month. A start that would breach it is refused before any money moves, and the refusal names the cap, the spend so far and when it resets. Leave it unset and only your balance bounds it.
One production at a time, per key
A key runs one production at a time by default. A storyboard waiting on you holds that slot — deliberate pressure, because the alternative is an agent starting a second paid video while the first is still unreviewed.
The balance is readable up front
An assistant can ask what a key can spend before it designs anything, so a shortfall turns up while the plan is still a conversation rather than after twenty briefs have been written.
Assistant work is priced from the same cost model as the app, with one channel-specific premium on longer, higher-tier jobs. The exact rule and when it applies are in the FAQ and the tool reference — and the quote always shows the final figure before anything is charged.
Getting connected
A key, a URL, and you're talking
Generate a key on the MCP and API Keys page in your account, bind it to your personal account or to one agency, and paste it into your client's config.
{
"mcpServers": {
"studiocut": {
"url": "https://app.studiocut.video/mcp",
"headers": { "Authorization": "Bearer YOUR-KEY-HERE" }
}
}
}Claude Desktop & Claude Code
Add the block above to your MCP config and the twelve tools appear in the next session.
Any MCP client
One HTTPS endpoint and a bearer token. If your agent framework can set a URL and a header, it can drive StudioCut.Video.
Agencies
A key is bound at creation to your personal account or to one agency, and can never switch. An assistant working a client's account cannot reach anything else. Need both? Make two keys.
Try before you build
Exercise the whole surface in your browser
We publish a standalone test console that calls every method and every tool against a live server. Point it at your key and watch the real requests and responses — useful for learning the flow, and for reproducing a bug before you open your own source.
MCP Test Console
Open source, runs locally, no build step. Get it on GitHub →
The tool reference
Every argument, every error code, and the order the calls have to happen in. Read the reference →
Credentials
Keys that expire on purpose
A credential that can spend money shouldn't live forever, and shouldn't quietly outlive the integration that used it. More on how the platform is secured on the security page.
Shown once
A key is displayed at creation and never again. Nothing in the platform can retrieve it afterwards — lose it and you make another.
Thirty days, maximum
Keys expire. You get an email a week before and a day before, so expiry is a scheduled event rather than a Monday-morning outage.
Rotation without downtime
Rotating issues the replacement immediately and keeps the previous key alive for a grace period, so you update a config file on your own schedule.
Bound at creation
Personal account or one agency — chosen when the key is made, immutable afterwards. There is no argument that widens what a key can see.
Unused keys are revoked
A key nobody has used for a fortnight is withdrawn after a warning. Any authenticated request resets the clock, so a live integration is never affected.
Every call is logged
Tool calls are recorded against the key that made them, so "what did the assistant actually do" is a question with an answer.
Plainly stated
What it doesn't do
So you can plan around it rather than discover it.
Publishing
No tool pushes a video to YouTube or anywhere else.
Saving brand presets
An assistant can brand a batch, but creating a reusable saved preset is a job for the Brand Library in the app.
Sign-in with OAuth
Authentication is bearer API keys today. OAuth is on the roadmap; the key path is the layer it will issue into.
FAQ
Frequently asked
Which AI assistants can connect?
Any client that speaks the Model Context Protocol over HTTP with a bearer token. Claude Desktop and Claude Code work today, as does any agent framework where you can set a URL and an Authorization header.
Can my assistant spend credits without asking me?
No. Starting a production requires a confirm token from a price quote that is only valid for five minutes, and if the price moves in between, the start is refused and re-quoted. You can also set a credit ceiling on each key, per day, week or month.
Can it publish to YouTube for me?
No. There is no publish tool. Pushing a video to a platform is irreversible and outward-facing, so it stays in the web app where a person does it deliberately.
Which credits does assistant work spend?
Your plan credits and your purchased credits. The get_balance tool reports what a given key can actually spend, along with any ceiling you set on it, so an assistant can pace itself instead of discovering the limit by hitting it.
Does my plan limit what the assistant can make?
Plan feature gates do not apply to assistant-driven production, and it does not consume the monthly video allowance you bought for the app. Platform limits still apply: storage, content policy, up to five extra languages per production, and the rate and concurrency limits on each key.
How long does an API key last?
Thirty days at most. You can rotate a key at any time, which issues the replacement immediately and keeps the old one working for a grace period, so a config file can be updated without downtime. Reminder emails go out a week and a day before a key expires.
Where do assistant-made videos show up?
On their own page in your account, kept separate from the main production grid so an automated batch never buries the work you are doing by hand. They are fully yours either way — same ownership, same downloads.
Ask for a video. Approve one storyboard. Get a video.
Free drafts, a quoted price before anything is charged, and one decision that is always yours.
Get Started Free