How to make video ads with Claude and Codex

Can Claude make video ads?

MCP (the Model Context Protocol) lets an AI app call another service's tools. Probably's server gives Claude the studio's: list your brand Spaces, prepare a video from a brief, start it once you agree, follow it until it's delivered, ask the crew for changes and share the result. The AI creative crew writes the script, casts the actor, films, voices, edits and reviews; Claude is the one you talk to.

What you need

• A Probably account; your first draft is free at 480p.
• Claude on the web or the desktop app (custom connectors are on every plan; Free allows one), Claude Code, or OpenAI Codex.
• The server's address, the same for every app:

https://studio.probably.co/api/mcp

Connect Claude on the web or the desktop app

1. Open Customize, then Connectors

   Choose + Add, then Add custom connector.

2. Name it and paste the address

   Call it Probably, enter https://studio.probably.co/api/mcp and choose Continue.

3. Sign in

   Keep the detected OAuth settings, choose Add, then Connect. Sign in to Probably and allow the connection.

On a Team or Enterprise plan an owner adds the connector under Organization settings, then Connectors; members then choose Connect on it under Customize, then Connectors. A connector added on the web works in the desktop app too.

Connect Claude Code

Add the server over HTTP:

claude mcp add --transport http probably https://studio.probably.co/api/mcp

Add --scope user to use it in every project. Then run /mcp inside Claude Code, choose probably and sign in in the browser it opens. The Claude Code quickstart also shows an API key for machines without a browser.

Connect OpenAI Codex

Add the server to ~/.codex/config.toml:

[mcp_servers.probably]
url = "https://studio.probably.co/api/mcp"

Then sign in:

codex mcp login probably

The Codex CLI, its IDE extension and the ChatGPT desktop app share this file. For CI, the Codex quickstart reads an API key from an environment variable instead.

What you choose when you connect

Probably's consent page asks which organization the app works in, optionally one Space, and what it may do:

• Read

  List Spaces, videos, personas and products; follow a video; get its links

• Make changes

  Save personas and products, message the crew, approve and share videos

• Start paid work

  Confirm a spend in the chat after you agree; without it, you confirm by a link in the studio

The app's access token lasts 60 minutes and renews itself for up to 90 days of use. Revoke it any time in Settings, then Connections; its next call is refused.

Make an ad from a chat

1. Find your brand

   Ask “List my Probably Spaces.” Claude calls list_spaces.

2. Brief the ad

   Ask for it as you would a person, such as the prompt below. Claude calls prepare_video and shows you the preview: whether it fits your credits, and whether your free draft covers it.

3. Say go

   Answer “Go ahead.” Claude calls confirm_spend, which starts exactly what the preview showed.

4. Follow it

   Ask “Tell me when it's delivered.” Claude calls wait_for_video until the video changes, and can relay the crew's questions.

5. Change, approve and share

   Ask for a change in words, approve the delivery, then ask for a link: get_media for a download, share_video for a lasting link.

Make a 15-second 9:16 selfie vlog for our Dewdrop face mist in the Dewdrop Space. A woman in her twenties sprays it on her commute and says why it lives in her bag. Show me the preview before you start.

That brief is one of the ad examples; each has its own brief to borrow.

How is spending confirmed?

• A preview waits 15 minutes for your answer, then expires unspent.
• Confirming the same preview twice starts the work once.
• Only the connection that prepared a spend can confirm it, and nothing written in a brief or a message can confirm it.
• See spend confirmations for the details.

The tools Claude gets

• list_spaces, list_videos

  Find your brands and their videos

• prepare_video

  Turn a brief into a draft and a cost preview

• confirm_spend

  Start the work a preview showed

• wait_for_video, get_video

  Follow a video and read what waits on you

• send_message

  Ask the crew for a change or answer them

• approve_delivery

  Accept the delivered video

• get_media, share_video

  Download it, or share a lasting link

Every tool is in the MCP tool reference, Connect an AI app covers ChatGPT, Cursor and VS Code too, and integrations lists every app Probably works with.