How to make video ads from Claude or ChatGPT with an MCP server
By Venkata Vamsi · Updated · 9 min read
Short answer
To make video ads from Claude or ChatGPT, add a video-ad MCP server as a custom connector, sign in once with OAuth, then ask for an ad in plain English. With Addy's server at use-addy.site/mcp, the assistant writes the hook and script, renders a 9:16 video in a few minutes and can schedule it to Reels, TikTok and Shorts.
When we built Addy's MCP server, the goal was simple: you should be able to say "make me a Reels ad for my site" in the AI chat you already use, and get a real video back, not a script you still have to film. This guide walks through the setup for each app, the prompts that work, what each tool actually does, and the limits you will hit.
We built the server ourselves, so this is less a feature tour and more a set of notes on how it behaves. If you have never heard of MCP: it is the Model Context Protocol, an open standard that lets assistants like Claude and ChatGPT call tools on other services. Addy's server is remote, which means there is nothing to install. You paste one URL and click Allow.
What do you need before you start?
Three things:
- An AI app that supports remote MCP servers. Claude (web, desktop and mobile), ChatGPT with developer mode, Claude Code, Codex, Cursor, VS Code with Copilot agent mode, Gemini CLI, Windsurf, Zed and Goose all work. Clients that only run local servers can use the open-source mcp-remote bridge.
- An Addy account. The 7-day trial needs no card and includes 150 credits, which covers one 20-second ad. Connecting over MCP works on every plan, including the trial.
- The server URL: https://use-addy.site/mcp
One plan note: a plain API key for headless, server-to-server agents is a Pro feature. If you are a person chatting with an assistant, you do not need one; OAuth sign-in handles it. The full reference lives on our MCP developer page.
How do you connect Addy to Claude?
On claude.ai or Claude Desktop:
- Open Settings, then Connectors, then Add custom connector.
- Name it Addy and paste https://use-addy.site/mcp. Click Add, then Connect.
- A browser window opens. Sign in to Addy and click Allow.
- In any chat, ask Claude to make an Addy ad. The tools show up under the + menu.
For Claude Code, run this in a terminal, then type /mcp inside Claude Code and choose addy, then Authenticate:
claude mcp add --transport http addy https://use-addy.site/mcp
There is also a shortcut. Paste this line into any agent and it will set itself up by reading our agent instructions page: "Connect Addy to yourself by following https://use-addy.site/agents, then make me a video ad for my product." Coding agents run the command themselves; chat apps tell you which three clicks to make.
How do you connect Addy to ChatGPT, Cursor and other apps?
Each app puts custom servers in a slightly different place. Here is the short version:
| App | Where to add it | Sign-in |
|---|---|---|
| ChatGPT | Settings, Apps and Connectors, Advanced settings: turn on Developer mode, then Create a connector with the URL and OAuth | Sign in to Addy, click Allow, then enable Addy from + and Developer mode in a new chat |
| Cursor | Add to Cursor button on our MCP page, or the URL in ~/.cursor/mcp.json | Opens the first time a tool runs |
| VS Code | Install button on our MCP page, or code --add-mcp with the URL | Prompted on first use |
| Codex | codex mcp add addy --url https://use-addy.site/mcp, then codex mcp login addy | Browser sign-in |
| Gemini CLI | gemini mcp add --transport http addy https://use-addy.site/mcp | Run /mcp auth addy |
| Local-only clients | npx -y mcp-remote https://use-addy.site/mcp as the command | Browser sign-in |
ChatGPT custom connectors need a Plus, Pro, Business, Enterprise or Edu plan, and developer mode has to be switched on first. If the Addy tools do not appear in a new chat, that toggle is almost always the reason.
Whichever app you use, the connection is tied to the Addy workspace you had selected when you clicked Allow. If you run several brands, switch workspaces in Addy and connect again.
What can you ask for once it is connected?
Plain English works. These are the kinds of prompts we test with:
- "Make an Addy ad for https://example.com. Something funny, claymation style. Show it to me when it's done."
- "Write 5 Addy concepts for my app, let me pick two, render them in Standard quality and schedule one per day."
- "Love it. Post it to my Instagram and TikTok tomorrow at 9am."
- "How many credits do I have left, and how many 20-second ads is that?"
- "Take the baby podcast format and make an ad about our free trial."
The concepts step is the one I would start with. It is free, it returns hooks and scripts you can read in the chat, and nothing renders until you pick one. That is the same swipe-to-approve rule the web app follows, just in conversation form. If you want ideas for formats to name, the format list shows all 16, from UGC talking head to chalkboard and car yapper.
One habit that helps: give the assistant your actual URL, not a description. The brand brief Addy writes from a real site, with its product names, tone and audience, produces noticeably more specific hooks than "a skincare brand".
What do the Addy MCP tools actually do?
Your assistant decides which tools to call, but it helps to know what is underneath:
| Tool | What it does | Spends credits? |
|---|---|---|
| write_ad_concepts | Writes several hooks and scripts to choose from | No |
| render_concept | Renders a concept you picked | Yes |
| create_ad | Learns the product, writes the script and starts the render in one call | Yes |
| get_ad | Waits for the render, returns the video URL, caption and hashtags | No |
| publish_ad | Posts or schedules to connected Instagram, TikTok, YouTube and more | No, but it posts publicly |
| list_ads | Your finished ads with URLs and views | No |
| list_formats | All 16 formats with their credit cost | No |
| list_social_accounts | Connected accounts, plus a link to connect more | No |
| get_brand, set_brand | See or re-learn the brand profile from a URL | No |
| get_account | Credits left, plan, where to top up | No |
| generate_video | Advanced: render straight from your own prompt | Yes |
| create_campaign | Advanced: a batch of concepts to review in the app | No |
A typical one-shot request is three calls: create_ad, then get_ad until the video is ready, then publish_ad if you asked it to post. If no social account is linked yet, publish_ad returns a link to connect one instead of failing, and the assistant can pick up where it left off after you connect.
Will the assistant spend credits or post without asking?
Only three tools spend credits or post: create_ad, render_concept and publish_ad (plus generate_video for advanced use). Claude and ChatGPT both ask you to approve tool calls by default; if you want to confirm every render, leave that setting on. The server's own instructions also tell assistants never to publish unless you explicitly ask.
Costs are the same as in the app. A 20-second Standard ad is 140 credits and HD is 280. On Starter that is about $1.35 and $2.70; the full breakdown is on our pricing page, and we explain where those dollars go in our render bill post. If a render fails, the credits come back automatically.
To disconnect, remove the connector in your AI app, or revoke it in Addy under API and MCP. Access ends immediately.
What are the limits of making ads over MCP?
Honest list, from our own testing:
- Renders take minutes, not seconds. Our first 8-second 720p test took about 4 minutes. Each get_ad call waits up to 45 seconds, so the assistant usually calls it several times in a row. If it stops early, ask again a minute later ("is my Addy ad ready?") and it will fetch the finished file.
- You may get a link, not a player. Depending on the app, the finished ad comes back as a URL rather than an inline video. Either way the file is also in your Addy library.
- The assistant can overreach. A vague prompt like "make me some ads" may lead it to render several at once. Ask for concepts first if you want control over spend.
- No timeline editing. You can ask for a new hook, format or length and re-render, but you cannot trim frames or move a caption by a few pixels from the chat. For that, download the file and use an editor.
- The video model has its own limits. It is weak at drawing on-screen text, logos and app screens, so Addy adds the hook text afterwards and avoids fake UI. Our notes on what AI video models get wrong cover the rest.
If you are a developer who wants the same thing without a chat in the loop, the REST API exposes brand analysis, concepts, rendering, jobs and scheduling directly.
Frequently asked questions
+What is an MCP server for video ads?
It is a service that exposes ad-making tools, like writing concepts, rendering a video and scheduling a post, over the Model Context Protocol. Once connected, assistants such as Claude or ChatGPT can call those tools when you ask for an ad in plain English.
+Can I make video ads with Claude?
Yes. Add https://use-addy.site/mcp as a custom connector in Claude's settings, sign in to Addy and click Allow, then ask Claude to make an Addy ad for your URL. Claude writes the concept with Addy, starts the render and returns the video link when it is done.
+Does the MCP server work with ChatGPT?
Yes, through ChatGPT's developer mode connectors, which are available on Plus, Pro, Business, Enterprise and Edu plans. Turn on developer mode, create a connector with the Addy URL and OAuth, then enable it from the + menu in a new chat.
+Do I need a paid Addy plan to use the MCP server?
No. Connecting is free on every plan, including the 7-day trial, which has 150 credits, enough for one 20-second ad. A static API key for headless agents requires Pro.
+Is it safe to let an AI assistant post ads for me?
Keep tool-call approval on in your AI app and it will ask before every render or post. Addy's server instructions tell assistants never to publish unless you ask, and you can revoke access at any time in Addy under API and MCP.