RendGet started

Blog · November 17, 2026

MCP Server for Video Ads: Make Ad Sets From Your AI Agent

An MCP server for video ads lets an AI agent like Claude, ChatGPT or Codex build ad creative by calling a video tool's functions: import your footage, write hooks, cut versions, resize for each placement, caption, and render the files. Rend's server at https://rend.ai/api/mcp does this from one connection, and it leaves every ad on an editable timeline instead of a pile of flat MP4s.

This post covers what to look for in an ad-focused server, the setup for each client, a prompt sequence that turns one product video into a test set, and where the approach falls short. For the general idea of editing over MCP, see MCP video editing.

What an ad server has to handle

A general editing server can cut a video, but ads ask for more because the work is mostly variation. You want the same body with six different openings, in three sizes, with the offer on screen in the right place for each platform, and you want to change the price on all eighteen when the promo ends.

So check a server against the jobs ads actually involve:

JobWhy it matters for adsRend
Hooks from one videoMost tests are hook testsSpins new hooks from your footage and brand
Sizes9:16, 1:1, 4:5 and 16:9 placementsMakes the sizes from one edit
Captions you can changePlenty of people scroll feeds with the sound offWord-timed, always editable, your font
Brand kitEvery ad should look like the same companyTemplates, voice, music bed and caption style reused
CopyHooks, scripts, headlinesrend-1, Rend's writing model with live search
Static and search adsA campaign rarely runs on video aloneImage ads and search ads from the same brand
ResearchKnowing what competitors runResearches the web, X posts and Meta ads
ReviewClients need to sign offReview link with notes drawn on the frame

A render-only server that takes a JSON template and returns a file can be fine for one fixed template. The cost shows up on round two, when every change means regenerating the template from scratch.

Connect the server

Rend's server is hosted, so there is nothing to install. It uses OAuth 2.1 with dynamic client registration and PKCE, which means most clients need only the URL.

  • Claude Code: claude mcp add --transport http rend https://rend.ai/api/mcp, then /mcp, pick rend, authenticate.
  • Codex: codex mcp add rend --url https://rend.ai/api/mcp, then codex mcp login rend.
  • Claude, Cursor and other clients: add a custom remote MCP server (connector) with the URL.
  • ChatGPT: add it as a custom MCP server from the Plugins page; our ChatGPT video editing guide has the current steps.

Your client opens a Rend page, you sign in and approve, and the agent can see Rend's tools. Clients that cannot sign in can use a personal token from Studio's agent connection settings, stored in the client's credential settings and never pasted into a prompt.

Worked example: one demo video to a 12-ad test

Here is a session for a made-up skincare brand with a 40-second founder demo filmed on a phone. The goal is a first test of four hooks across three sizes.

Prompt 1, the brief.

Make a new Rend project called "Serum launch test". Import this video: [direct HTTPS link]. It's a founder demo for a vitamin C serum, $38, 20% off with code GLOW20 through Nov 30. Build a 20-second ad from the strongest product moments. Vertical first, 30 fps.

The agent creates the project, imports the clip, transcribes it, and cuts a 20-second body on exact transcript words. It returns a Studio link where you can see the cut.

Prompt 2, the hooks.

Write four different opening 3 seconds for this ad: one that starts on the problem (dull skin in winter), one on the result, one on the price, one on a customer-style objection ("I've tried vitamin C before"). Keep the same body after the hook. Show me the hook lines before you build them.

Ask to see the lines first. Reading four hooks takes thirty seconds, and it is cheaper to reject a weak line in text than after a render.

Prompt 3, the sizes and offer.

Build all four in 9:16, 4:5 and 1:1. Captions in our brand font, kept out of the platform UI zones. Put "20% off, code GLOW20" on screen for the last 4 seconds. Name files by hook and size.

That is twelve ads. Check the framing of one file per size with the video ad sizes checker before you trust the rest.

Prompt 4, the share.

Render all twelve and send me the review link.

Your team reviews on one link, drawing on the frame where something is off. If the promo code changes, the fix is a text change on the timeline the agent made, then a new render. Our guide to ad video review has a checklist for that pass.

Prompts that save time on ad work

These work in any client once the server is connected.

  • "Research the Meta ads running for [competitor] this month and list their hooks and offers. Don't copy them; tell me what angles are missing."
  • "Make three static image ads and four search ad headlines from this same brand and offer."
  • "The client wants the logo smaller on every version. Change it once and re-render the set."
  • "Cut a 6-second version of each hook for placements that need short ads."
  • "Use our brand kit for every ad in this project."

For bigger runs, our guide to mass ad creation shows how to plan a test matrix of 100 variations without drowning in files.

Running the ads

When the set is done you have two paths. Download the files and upload them to your own ad accounts, or hand the set to Rend Boost, where Rend runs them live on major ad networks and you pay per click. Boost is in early access.

Where this falls short

An agent is quick at variation and slow at taste. It will happily give you twelve versions of a weak idea. Spend your attention on the brief and on the hook lines in prompt 2, because everything downstream copies them.

Your source footage still sets the ceiling. A shaky, dim demo can be cut tighter and captioned well, but it will not look like a studio shoot. If you have no footage, Rend can generate shots and b-roll, but read our post on making AI videos that don't look like AI before you build a whole ad from generated clips.

MCP clients also vary in how they handle files. If yours cannot pass chat attachments to the server, give it direct HTTPS links to your media instead.

FAQ

What is an MCP server for video ads?

It is a server that exposes ad-making tools (import, cut, hook writing, resizing, captions, rendering) over the Model Context Protocol so an AI agent can call them. You talk to the agent; the server does the video work.

Can Claude make video ads?

Claude cannot edit video by itself, but connected to a video server it can direct the whole job. With Rend connected it can build hooks, sizes and captions and return rendered files plus an editable project.

Do I need a video to start?

It helps. Rend can also start from a product page or a TikTok that worked, and it can generate shots, but real footage of your product usually makes a stronger ad.

How much does it cost?

With Rend, editing is free and you pay when it renders or generates something. There are monthly credit plans and pay-as-you-go; see pricing.

Can I edit the ads myself afterward?

Yes. Every ad opens in Rend Studio on the same timeline the agent built, so you can move a cut or change a caption by hand.

To try it, create a free Rend account, connect your agent with the command for your client, and give it one product video and your offer.

Give Rend the job.

Editing is free. You pay only when Rend renders or generates something.