Best AI Video MCP Servers for Making Ads: 15 Compared, Beyond Higgsfield's List

Ten servers made Higgsfield's list. No ad-first tool did.

Higgsfield's numbered list, which puts Higgsfield first, went up on October 20, 2026. It names ten AI video MCP servers for Claude, ChatGPT and Cursor, and none of the ten is an ad maker such as Arcads, Creatify, MakeUGC, TopView or Novoads. All five document an MCP server of their own. This guide adds them.

An MCP server is how an AI agent reaches a product's tools. You paste one URL or run one command in Claude, ChatGPT, Cursor or Claude Code, sign in, and the agent can start a render, check on it and fetch the file. For ads, three things separate one server from another: what it can make, how it handles a render that takes minutes, and whose balance each run spends.

Below are all 15, Higgsfield's ten plus the five ad makers, compared on those points. We have not used the rival servers hands-on: their facts come from each vendor's own pages, read on October 23, 2026, or dated as the page is. The one server we ran is ours, and that 6-credit run from Claude Code is shown call by call further down.

Quick pick: which AI video MCP server fits which ad

Quick pick by use case

Each pick rests on the vendor's own MCP page:

• Talking-actor ads from a chat: Arcads, whose tools include talking actors, editing and audio.
• Ads cut from a product page: Creatify, whose preset reads "Product page in, three 15-second Meta cuts out."
• A product in a model's hand: TopView, whose sample image prompt puts an uploaded product in a man's hand.
• Avatar video at batch scale: HeyGen, which takes up to 100 videos in one call.
• UGC flows on a MakeUGC plan: MakeUGC, connected with its Platform API key.
• A Runway or Higgsfield plan you already pay for: both bill a run to that plan's credits.
• A price before every render: Novoads, where estimate_cost spends nothing, or Vivideo, which lists estimate_credits.
• Raw models and your own queue: fal.ai.
• Your own weights on your own hardware: mcp-video-gen.

All 15 servers side by side

Clients are the ones each vendor documents, and billing is what its page says a run draws on.

Server	Documented clients	Sign-in	Makes for an ad	Billing
Arcads	Claude, ChatGPT, any MCP app	Arcads account	Talking actors, video, images	Arcads credits
Creatify	Claude, ChatGPT, any client	OAuth	Persona reviews, product-page cuts	Plan credits
MakeUGC	Claude Code, Claude Desktop	Platform API key	Flows and models, listed in Claude	Studio credits
TopView	Claude, ChatGPT, Cursor	TopView account	UGC videos, product visuals	Unlimited plan excluded
Novoads	Claude, Claude Code, any OAuth client	OAuth, no key	Video clips, ad images	Plan credits
Higgsfield	Claude, ChatGPT, Cursor, CLI	No key, paid plan	Creator-led product videos	Higgsfield credits
Runway	ChatGPT, Claude, Cursor, Grok Bot, Replit	Runway account	UGC-style ads, models by plan	Runway credits
HeyGen	Claude, Cursor, Gemini CLI, ChatGPT	OAuth, no key	Avatar videos, batches of 100	Premium plan credits
fal.ai	ChatGPT, Claude, Cursor, Codex CLI	OAuth relay	Raw model runs	fal credits
Dora	ChatGPT, Claude, Cursor, Antigravity	OAuth 2.1 or key	Text- and image-to-video	Dora coins
Vivideo	Claude, Cursor, any client	API key	Video, avatars, voices	Usage-based
AITuber	Claude, ChatGPT, Cursor, more	Sign-in or key	Scripted videos, auto-published	AITuber credits
OpenArt	Cursor, Claude, ChatGPT	Browser sign-in	Ad videos from chat	openart.ai credits
Imagine.art	Eleven via installer, ChatGPT	OAuth, no keys	Presenter ads, UGC avatars	Plan credits
mcp-video-gen	Self-hosted	Portainer deploy	ComfyUI or Blender jobs	Your own hardware

How we compared them

Each server was checked on six points, taken from its own MCP page, help center or developer docs:

1. The endpoint or install route.
2. The clients the vendor documents.
3. Sign-in: OAuth in the browser, or an API key.
4. What it can make for an ad.
5. How it handles a render that outlives the tool call.
6. Whose balance a run spends.

Higgsfield's list sets its own rule: servers with their own MCP docs for Claude, ChatGPT or Cursor, plus one open-source option to self-host, with facts verified October 17, 2026. Only its first slot comes with a reason, that one connector covers "a multi-model catalog together with characters, audio and ad workflows". Slots 2 to 10 carry no stated criterion, so the order here is ours.

Four ad makers with an MCP server of their own

Arcads, Creatify, MakeUGC and TopView sell ad production first. Novoads, the fifth, gets its own section below because it is the one we ran.

Arcads: talking actors and ad research from Claude

Arcads' help article of October 14, 2026 gives the connector URL as https://mcp.arcads.ai and says the MCP works "from inside Claude, ChatGPT or any MCP app". You sign in with your Arcads account, and Claude then calls Arcads' tools: "video and image models, talking actors, editing, audio, presets, competitor ads, social research and creator search".

Generations "come out of your Arcads credits at the same cost as in the app", and the article says every plan includes the MCP, "including the free trial". A second, smaller server, the Arcads lite MCP, carries about 43 tools for reading, uploads and analysis. Our walkthrough of Arcads' MCP and its community Claude Code repo covers the setup.

Creatify: product pages in, Meta cuts out

Creatify's MCP page, read on October 23, 2026, says its server at https://api.creatify.ai/mcp "works with any MCP client", with skills aimed at Claude and ChatGPT. Setup is standard MCP over HTTP with OAuth, signed in with your Creatify account.

Its presets read like an ad brief:

• "Five creator-style reviews, one per buyer persona."
• "Product page in, three 15-second Meta cuts out."
• One source portrait turned into a reusable on-brand creator.

Creatify also puts a number on time. A video usually takes 5 to 15 minutes, runs are async, and the agent polls the task instead of blocking. Generations use the same plan credits as the platform, and each turn reports the credits it used. Our Creatify review covers the product behind the connector.

MakeUGC: a Platform API key, even on the OAuth route

MakeUGC's platform docs point Claude at https://mcp.makeugc.ai and document the connection for Claude Code and Claude Desktop. Its docs ask for a key: you need "a Platform API key" from its settings "and an active subscription", and the OAuth route still has you paste that key on the consent page.

For the tool list, the MCP section sends you to Claude: "ask Claude to list MakeUGC flows or models" once you are connected. The same key drives MakeUGC's REST platform API, which documents product-in-hand videos and avatar videos from a script. Its pricing page says API and MCP access come with every plan at no extra cost and draw on the studio's credit balance. Our MakeUGC review covers the studio itself.

TopView: UGC videos on a canvas, but not on Unlimited

TopView's MCP page gives https://mcp.topview.ai/mcp, names Claude, ChatGPT and Cursor, and has you sign in with your TopView account once. TopView says the server orchestrates image and video models "to create storyboards, product visuals, UGC-style videos, ad edits, and multiple creative versions on one infinite canvas", and one of its sample image prompts asks the agent to put an uploaded product in a man's hand.

The catch sits in its FAQ: the Unlimited plan "cannot be used with Topview MCP, Plugin, or Skill". Our TopView review covers the rest of the product.

Higgsfield, Runway and HeyGen: the top of the list

Higgsfield: one connector, plus a CLI for coding agents

Higgsfield's help center, last modified September 24, 2026, says it connects to Claude, ChatGPT, Cursor and other MCP agents "with no API key". Claude takes https://mcp.higgsfield.ai/mcp as a custom connector, while Claude Code, OpenClaw and Hermes install the Higgsfield CLI instead. An active paid subscription is required, and every generation through MCP, the ChatGPT plugin or the CLI "deducts credits at standard rates".

For ads, its UGC Factory skills make creator-led product videos, including Product Review UGC. The list describes long renders as "Job IDs with a long-poll tool that tracks up to twelve jobs", and the help center says results typically appear within 60 seconds of a job completing.

One point Higgsfield states two ways. Its October 20 list counts Marketing Studio among what the connector covers. Its help article on Marketing Studio, last modified September 1, 2026, answers "Can I use Marketing Studio through MCP or Supercomputer?" with "Not yet." Our Higgsfield MCP explainer goes through its billing rules.

Runway: your Runway plan, inside the agent

Runway lists ChatGPT, Claude.ai, Claude Desktop, Grok Bot, Cursor and Replit. ChatGPT, Grok Bot and Cursor take a plugin; everywhere else you add https://mcp.runwayml.com/mcp and sign in with your Runway account, with no API key. Models follow the plan, and the page names Seedance 2.5, Kling, Gen-4.5, Veo, GPT Image 2 and Nano Banana Pro among them.

Its sample prompts are ads, one of them a "15-second TikTok Shop style UGC ad for this chair". MCP uses your Runway credits, just like the app.

HeyGen: avatar video, and batches of 100

HeyGen's remote server is at https://mcp.heygen.com/mcp/v1/. Its developer docs name Claude, Cursor, Gemini CLI and Manus, and ChatGPT gets a plugin connected with OAuth. Sign-in is OAuth tied to your HeyGen account, with no API key.

Its create_video tool makes an avatar video, in which "a HeyGen avatar or an image speaks your script or audio with lip-sync", and create_video_agent makes a whole video from one prompt. This is the server built for volume: create_video_batch submits up to 100 videos in one request and returns a batch id at once, and get_video reports each video's status. Renders use the premium credits in your existing plan, with "no separate API charges or additional billing".

The other seven: raw models, coins and self-hosting

fal.ai: the model catalog, billed per job

fal's MCP docs, modified October 22, 2026, put an OAuth relay at https://mcp.fal.ai/mcp-relay for ChatGPT, Claude, Claude Code, Codex CLI, Cursor and Grok Bot. The server lets an assistant "find models, check schemas and pricing, upload inputs, and run generations", and generations use fal credits.

Our reading: it is a model runner, so the actor, the voice and the ad format are yours to put together. For video, fal says to start with submit_job and "do not resubmit to check progress", because another call creates a new billable job.

Dora, Vivideo and AITuber

• Dora serves ChatGPT, Claude, Cursor, Google Antigravity and other clients at one endpoint, https://doravideo.com/mcp, with OAuth 2.1 or a static API key. generate_video does text-to-video, image-to-video or motion transfer and returns a job id, and check_job returns the file. Runs draw on one coin balance shared with Dora's website and iPhone app.
• Vivideo's page, modified July 19, 2026, ships the server as an npm package, npx -y @vivideo/mcp, for Claude, Cursor and any MCP client, secured with an API key on usage-based pricing. It covers text-to-video, image-to-video, avatars and voices across 30+ models, and its tools include estimate_credits and wait_for_video.
• AITuber runs at mcp.aituber.app for Claude, ChatGPT, Cursor and other clients. You sign in, or pass an API key as a header when nobody can. It writes scripts, browses 1,300+ voices, generates videos and publishes to YouTube, TikTok and Instagram. The server is free, and videos use AITuber credits.

OpenArt, Imagine.art and mcp-video-gen

• OpenArt's Cursor page gives https://mcp.openart.ai/mcp, which also works in Claude and ChatGPT. You approve access in the browser, with no API key, and the setup ends by asking Cursor "to make an ad video with OpenArt". Generations use openart.ai credits, and MCP comes with every plan from Starter up.
• Imagine.art serves https://mcp.imagine.art, with an installer that sets it up in eleven clients, from Claude and Cursor to Gemini CLI, plus a ChatGPT plugin, and sign-in is OAuth. Its Ad Studio makes "unboxing, review, tutorial and testimonial ads with a presenter on camera", other tools cover product ads and UGC avatars, and clips run five or ten seconds by model. Jobs run async, on the platform's own credits.
• mcp-video-gen is an Apache-2.0 GitHub project, not a company: a self-hosted server, deployed through Portainer, in front of local backends such as ComfyUI and Blender. It ships no generation workflows or agent skills, so it makes whatever your backends make.

How Novoads' connector ran one ad brief, call by call

Setup: one URL, OAuth and two permissions

The Novoads connector is at https://novoads.ai/api/mcp. Our connector page says it works with anything that speaks MCP over Streamable HTTP with OAuth, and names Claude and Claude Code as the two our team uses daily. Sign-in is OAuth with no API key: approve two permissions, then ask for an ad. In Claude Code it takes one command:

claude mcp add --transport http novoads https://novoads.ai/api/mcp

Screenshot: novoads.ai/mcp, the three steps that add the Novoads connector to Claude, taken 2026-10-20

The run: estimate, generate, poll

On October 20, 2026, a Claude Code session asked the connector for one UGC clip: a woman in a bright kitchen holds a plain sage-green insulated bottle up to her phone camera and says "I filled this with ice yesterday morning. Still ice. I'm not kidding." Three tools did the work:

1. estimate_cost quoted 6 credits and spent nothing.
2. generate_video, set to Seedance 2.0 Mini at 6 seconds, 9:16 and 720p with audio, returned a job id at once instead of holding the connection open.
3. get_generation said queued on the first poll and succeeded on the second, with the MP4.

The job was created at 11:00:38 UTC and completed at 11:02:33. Novoads debited 6 credits, and the spoken line came back word for word. One miss: the model added a small emblem near the bottle's base although the prompt asked for no logo, so a branded product should go in as an uploaded photo, and every frame gets a look before the ad runs.

Seedance 2.0 Mini · prompt: a woman in her kitchen holds a sage-green insulated bottle to her phone camera and says the ice is still frozen a day later · 6 credits · rendered 2026-10-20

The full prompt and every response from that session are in our Claude ads skill walkthrough.

What a run costs, and what else the connector makes

At plan prices on October 23, 2026, 6 credits is 45 cents on Starter, which is $15 a month for 200 credits, and about 29 cents on Plus, which is $49 a month for 1,000 credits. Every plan is on the pricing page, and a connector run spends credits from your plan on the same schedule the app charges.

The connector page lists 10 tools, 5 of which spend credits, while the rest read, price and report. generate_image makes a static ad image. generate_video runs Seedance 2.0 by default, and a default-length Seedance 2.0 video is 23 credits. For a branded product, the photo goes up first through upload_asset.

What to check before an agent spends your credits

Whose balance a run spends

Where the cost of a run lands:

• Your plan's own credits: Higgsfield, Runway, Arcads, Creatify, HeyGen's premium credits, Imagine.art, OpenArt, MakeUGC's studio balance and Novoads.
• A separate unit: fal credits, Dora coins, Vivideo's usage-based pricing and AITuber credits.
• Your own machine: mcp-video-gen.

Two plan rules can stop a run before it starts. Higgsfield's connector needs an active paid subscription, and TopView's Unlimited plan cannot be used through its MCP at all. Check both before you pick a plan for agent work.

What happens when a render outlives the call

A video render can take minutes, longer than an agent waits on one tool call, so the safe pattern is a job id now and a poll later:

• Described: Novoads, Creatify, HeyGen, fal.ai, Dora, Vivideo, Imagine.art and Higgsfield each describe a job id, an async run or a wait tool, and MakeUGC's REST API asks callers to poll a status endpoint.
• Not described on the pages we read: Runway, Arcads, TopView and OpenArt.
• The times on record: Creatify's 5 to 15 minutes, and about 4 to 7 minutes in the description of Novoads' own get_generation tool. Our run took under two.

The failure to avoid is a retry that pays twice. fal warns that resubmitting to check progress creates a new billable job. Novoads' generate_video takes an idempotency key, and the same key within 24 hours returns the first submission instead of rendering and charging again.

Keys, OAuth and servers that only read

Most of the 15 sign in through the browser, so no key sits in a config file. The exceptions matter for a shared or headless agent:

• A key: MakeUGC wants its Platform API key even on the OAuth route, and Vivideo's server is secured with an API key.
• Either: Dora takes OAuth 2.1 or a static API key, and AITuber takes a key in a header when nobody can sign in.
• Self-hosted: mcp-video-gen runs on your own Portainer setup.

Not every MCP server near ads makes them. Arcads' lite server reads and analyzes, and our breakdown of Google's official Ads MCP server covers what an agent connector can and cannot do inside an ad account.

Pick the server by the ad, not the model count

A list of AI video MCP servers is mostly a list of model catalogs with a connector in front. For ads, the connector is the part that decides the outcome: whether it knows actors, products and ad formats, whether it hands back a job id when a render takes minutes, and whether it prices a run before it spends your credits. On those three questions, the five ad makers belong on any list of the best AI video MCP servers, ours included.

Choose the server by the ad you need to ship, then let one priced run tell you the rest.