Your Desk, from Claude or ChatGPT. Approval stays with you.

Yes, you can connect Claude or ChatGPT to Corantr Media Desk. Add a custom connector with the address https://corantr.com/mcp, sign in to the Desk when asked, and press Allow. The assistant can then list your sites, find fresh stories, start pieces and export approved ones. It can't approve or publish: that stays with a person on the Desk.

The Desk has an MCP server, which lets AI assistants use the Desk on your behalf. In practice, that means you can ask for things in plain words in a chat window, and the assistant carries them out on your Desk. If you already spend part of your working day in Claude or ChatGPT, we'd connect it. You stop switching windows to find stories, start pieces and fetch finished copy.

What the assistant can do on your Desk

Once connected, the assistant works with the same Desk you use in the browser. It can do these things:

• List your sites.
• Go through your ideas: list them, approve or turn them down with a reason, and write one.
• Find fresh stories from your news feeds.
• Start a piece from links or a brief. That can be a news piece, an article, social posts or a video script.
• Check how a piece is getting on, and read pieces.
• Export approved pieces as Markdown, HTML or text.

Starting a piece uses your credits, just as it does on the Desk. The assistant takes the same route you would, so a piece started from a chat window costs what it would cost if you'd pressed the button yourself. It's worth deciding what you want before you ask, the same as you would on the Desk.

If you're one of the agencies running several client sites, asking for a list of sites is a sensible first request. It shows you which sites the connection can reach before you start anything.

Reading pieces is more useful than it sounds. You can check a piece's wording in the chat window while you work on something else, then go to the Desk when you're ready to deal with it. For export, HTML suits most website editors; Markdown or plain text work for anything else.

How to connect in Claude or ChatGPT

The steps are the same in both assistants.

1. In Claude or ChatGPT, add a custom connector.
2. Give it the address https://corantr.com/mcp.
3. When asked, sign in to Corantr Media Desk with your usual account.
4. Press Allow.

There's no token to copy. Signing in and pressing Allow is the whole handshake. If you're asked to paste a key into Claude or ChatGPT, something else is going on: stop and check the address.

In ChatGPT, your workspace may need developer mode turned on before custom connectors appear. If you can't find the option to add one, that's the first place to look. Whoever manages your ChatGPT workspace will know whether it's switched on.

Using Claude Code, Cursor or another tool

Developer tools connect differently. For Claude Code, Cursor and other tools, you make a personal API token on the Desk under Settings → AI assistants, then give that token to the tool.

Any other tool that can connect to an MCP server with a token can use the same route. Treat the token like a password. It's personal to you, so don't share one between colleagues: each person who wants to connect a tool makes their own.

The two routes reach the same Desk. A custom connector suits Claude and ChatGPT; a personal token suits developer tools. Pick the one that matches where you work.

What the assistant can't do

The assistant can't approve or publish anything. Every piece it starts still waits on the Desk for a person.

The connection also works inside the organisation you're in on the Desk, with your role. So the assistant can't do more than you could do there yourself. If your role doesn't let you see a client's site on the Desk, the assistant can't see it either. If you work across more than one organisation, it works in the one you're signed into. When a site seems to be missing from the list, check where you're signed in before anything else.

This is why we're comfortable recommending the connection. The assistant takes on the fetching and starting. Every piece is still written from a fact sheet taken from the source and checked claim by claim, and the decision to publish sits where it always has.

A morning's work from a chat window

Here's how a morning might run with the connection in place. It's an illustration of the steps, not a real customer's day.

You open Claude and ask it to list your sites. It returns the sites in the organisation you're signed into.

You pick one and ask for fresh stories from that site's feeds. The assistant brings back what the feeds have found. You read the list and choose a story that affects your readers.

You paste in the link to that story and ask the assistant to start a news piece from it. Then you write two lines describing an event you're running and ask for social posts from that brief. Two requests, two pieces started.

You get on with other work. Later, you ask how the pieces are getting on, and read the news piece once it's ready. One sentence about the event needs a change, so you make a note of it.

Now you go to the Desk. You read the piece, make your edit, and approve it. The social posts get the same treatment.

Back in the chat window, you ask the assistant to export the approved news piece as HTML. You copy it into your site and publish it yourself.

Most of the steps happened in one window. The step that mattered happened on the Desk, with you.

Switching it off

You can disconnect the assistant at any time under Settings → AI assistants on the Desk. That's the same page where developer tools get their personal tokens, so it's the one place to go for every assistant connection on your account.

If you already work in Claude or ChatGPT, connect it and try the morning above on one of your own sites. Approval stays with you on the Desk, and you can switch the connection off whenever you choose.