Video MCP connector: let your AI assistant work with your DCAST video library

If you keep your videos and live broadcasts in DCAST, you can let an AI assistant work with that library instead of clicking through every screen yourself. The DCAST connector is an MCP server for video that connects an AI assistant such as Claude to your DCAST account, so you can ask in plain language what is in your library and have simple changes made for you. This article explains what you are connecting, which access you give, which doors stay closed to an assistant and how you take the connection away again. You also get a short set of habits that keep you in charge of your library.

What the video MCP connector is

MCP is an open way for an AI assistant to use tools that live outside the chat window. The DCAST connector speaks that language on one side and works with your account on the other, and Claude on the web, Claude Desktop, Claude Code or any other MCP client can use it. You sign in with the DCAST account whose library you want to work on, so the assistant sees and changes the content of that account and not the content of anyone else's. Think of it as a new colleague who logs in with a limited badge.

The connector adds no powers of its own. A tool can do exactly what the same credential can do through the DCAST partner API, so an assistant is never more capable than the access behind it. That does not mean the rules are the ones you know from your dashboard. An assistant works under its own grant, which holds a role and a set of areas, and DCAST checks every call against that grant before anything happens. Read the access you approve with that in mind, because it is a separate decision from the rights you hold when you work in the dashboard yourself.

Connecting your assistant and choosing access

Connecting starts in the assistant. In Claude you add a custom connector and sign in, and the sign-in page lives on dcast.tv or dcast.pro, depending on your account. After you sign in you reach a consent screen. Read it before you click, because a connection approved in a hurry is easy to regret. What the assistant can do still follows the limits of your plan, so do not be surprised when a request above those limits is stopped.

On the consent screen you decide how much to allow, and it offers three choices: view only, manage streams and videos, or customize manually, where you pick the level for each area yourself. Start with view only when you want answers, such as which videos are still private, because then nothing can be changed. Choose the manage option for a session in which you want videos and playlists edited, and use the manual option when the job touches only one or two areas of your account.

Do not treat the manage option as harmless. Besides running streams and restreams and editing playlists, it includes deleting videos. If your job is only to fix titles, open the manual option and keep the level for videos below delete yourself. Levels also differ between areas: for videos, deleting is a separate level above changing, while an area such as playlists has a read level and a write level. Before you approve, ask which parts of your account the job really touches, and keep the grant as narrow as the task allows.

Live broadcasts bring a second concern. Stream keys stay hidden unless you ask for one, which means a casual question about your broadcasts does not put a secret into a chat transcript. Even so, treat any chat that contains a stream key as sensitive and do not paste it anywhere else.

What stays closed to an assistant

Some actions are closed to an assistant whatever you click on the consent screen. Money, such as payouts and checkout, API keys and changes to team membership belong to you in the dashboard, and an assistant that tries them is refused. If you ask an assistant to change your pricing or invite a colleague, expect it to send you back to do that yourself, and plan for that instead of looking for a workaround.

A request that goes beyond the areas of your grant is refused as well. Treat such a refusal as a prompt to decide, not as an obstacle to push through. Ask yourself whether the job justifies a wider grant, and if it does not, finish the work in the dashboard instead.

Removing the connector and why the removal holds

When you remove the connector in your DCAST API keys, it leaves your list of keys. Behind the scenes DCAST keeps a switched-off record of the connection instead of deleting it, and that choice is what makes the removal final. A token the connector received before the removal is refused from that moment, even if it had not expired yet. Without the record, a still-valid token could have created the connection again on its next call.

A newer token cannot reopen the removed connection either. To connect again, you approve the access in DCAST once more, which makes reconnecting a deliberate step. Keep that in mind when you swap one assistant for another: removing is quick, but a fresh connection needs you at the consent screen.

Removing the connector also closes the permission requests it still had open, so an old approval cannot widen a new connection. This is easy to overlook. If the assistant asked for more access earlier and you never answered, that request would otherwise still be waiting when you reconnect. After a removal, check that the entry is gone from your list of keys, then ask the assistant for one harmless action and expect it to fail.

Another account cannot remove your connector, because such a request is answered as not found and your connection stays as it was. That has two consequences for you. A removal you did not make is a warning sign worth investigating, and you have no reason to share your login so that a colleague can tidy up keys. Invite that person with their own access instead.

Habits that keep you in charge

Ask your assistant to tell you what it is about to change before it changes it, so you stay in charge of your library. A good request names the videos, the field and the new value, and asks for a summary before anything is applied. For a bulk job such as renaming many videos, start with two or three, read the result and only then let the assistant continue.

Remove the connection you no longer use, so no forgotten assistant keeps a way into your account. Open your API keys now and then, look for entries from assistants you stopped using and remove them. A tidy list also makes the day you must remove one in a hurry less stressful, because you already know what each entry is.

If you suspect that someone else used the connection, remove it first and connect again afterwards with a fresh approval. Do not spend time investigating while the connection is still open. Once it is removed, tokens issued before the removal are refused, so you can look into what happened calmly and restore anything that should not have changed.

After a session, look over what the assistant changed in your library, the same way you would check the work of a new colleague. Compare titles, visibility and descriptions with what you asked for, and open a few videos in the player to be sure they still look right. Do this even when the assistant reported that everything went well, because a summary is not the same as a check you made yourself.

Checking failures and knowing when the connector fits

If the assistant says it cannot reach your library, check three things in order. First, look at whether the connection still appears in your list of keys, because a removed connection stays removed until you approve it again. Second, ask whether the access you granted covers the request, since a call above the granted areas is refused. Third, consider your plan, because a feature that your plan does not include can stop a call as well. Decide whether the job justifies a wider grant before you ask again.

A refused call is not necessarily a fault, because it often shows that the limits you set are working. A message saying the connection is no longer active has more than one possible cause, and a removal is only one of them, since an inactive or expired grant gets the same answer. Check your list of keys first, and reconnect only if you meant to keep the assistant, approving the smallest access that does the job.

The connector suits routine library chores: finding videos, tidying titles, checking the status of a broadcast and answering questions about what you have. It is less suited to work where one wrong step is costly, such as removing a large part of your library, so do that in the dashboard where you see every item. If your own website must add videos or create broadcasts with nobody present, use a partner API key kept on your server instead, because that is a different way in from an assistant that signs in through a consent screen.