AI video generation for agents is one of those ideas that sounds like a demo-day gimmick until you run it inside a real workflow. Popcraft, which launched on Product Hunt on 29 September 2026, takes a route every other video studio has ignored: instead of an app you open, it is an entire video platform exposed as an MCP server. Paste one URL into Claude, ChatGPT, Cursor or Claude Code, and your agent can plan, cost-preview and generate video, images, music and sound effects with Seedance 2.5, Veo 3.1, Kling 3.0, Nano Banana Pro and 20+ more models — without you ever touching a timeline.
The verdict up front: if your team already lives inside an AI agent, Popcraft is currently the most natural way to turn a chat brief into finished video. OAuth sign-in instead of API keys, a cost preview before the agent spends a single credit, and 100 free credits to start make the trial genuinely free. If you just want a manual studio for hand-crafted clips, a dashboard tool will suit you better. For everyone else — especially agencies and marketing teams already briefing Claude or Cursor — this one deserves a look.
What Popcraft actually does
The core insight, straight from co-founder Eric’s launch post on Product Hunt: Popcraft started as “one studio for the best AI video, image and audio models,” then the team noticed they had stopped opening the app. They were briefing Claude and Cursor and wanting the output to simply appear. So they shipped Popcraft MCP — the whole platform as an MCP server.
The mechanics are refreshingly simple:
- One endpoint:
https://popcraft.ai/api/mcp— paste it into any MCP-capable host. No API key; you sign in with OAuth and the token auto-refreshes. - 27 tools: explore models, upload reference images, preview cost, generate video/image/audio/3D, and upscale.
- Works everywhere agents work: Claude, Claude Code, Cursor, ChatGPT, Codex — anything that speaks MCP.
- Cost preview: video is the expensive medium, so the agent shows you what a generation will cost in credits before it spends anything.
There is also a less obvious but important piece: production workflows served live. Ask the agent for “a 15-second vertical UGC ad for this product” and it does not fire off a prompt and hope. It follows a recipe the Popcraft team built in — intake, creator identity, storyboard, a realism pass, then one Seedance clip with native audio. That recipe layer is what separates “an API with a chat window” from something you can trust with client work.
AI video generation for agents: the model roster
Popcraft aggregates the models that are actually winning in late 2026 rather than chasing every new release. The launch roster includes Seedance 2.5, Google Omni 1.1, Kling 3.0 Omni, Nano Banana Pro, GPT Image 2.5 and ElevenLabs, plus Veo 3.1 — 20+ models across video, image, audio, 3D and upscaling. The maker is candid about what each is good at today: UGC product ads generated from a single product photo, YouTube thumbnails with identity lock, 30-second single-take Seedance 2.5 clips, product image sets, and voiceover plus music.
A quick orientation on the three headline video models, drawn from how the wider market describes them:
- Seedance 2.5 — the long-take specialist. Up to 30 seconds in one generation with native audio, and it can fold in many references (images, audio, even video). Ideal when one continuous scene matters more than maximum polish.
- Veo 3.1 — Google’s flagship for short, polished shots: 8-second clips, native synchronised audio, 4K output, and first-and-last-frame control. The pick for the hero shot.
- Kling 3.0 — the multi-shot model. A single generation can contain several cuts and camera setups while keeping a character or product consistent — genuinely useful for ads that need more than one angle.
For a deeper model-by-model comparison of the image-to-video generation landscape, this Veo 3.1 vs Kling 3.0 vs Runway roundup is a solid reference, and the Kling AI breakdown covers where Kling wins in practice.
Pricing: what a minute of video actually costs
Popcraft runs on credits, and the honest numbers are public. You start with 100 free credits — no card required. Generation is billed in credits per second of output at your chosen resolution. Popcraft’s own pricing notes for Seedance 2.5 put the list rate at 42 credits per second at 720p, with 480p at 24 credits per second as the cheaper way to iterate before a final pass. So a 15-second 720p clip costs roughly 630 credits at list rate.
Paid plans cut that rate rather than just adding credits: Pro takes 20% off the rate on monthly billing (40% on annual), and Max takes 40% off monthly (60% on annual), applied automatically to every generation. Pro and Max also include a daily allowance of Seedance 2.0 Fast & Mini generations, which covers quick iterations without touching your credit balance. Popcraft publishes the live per-model rates on its pricing page, so check the current dollar figures there before committing — this review will not quote dollar prices it cannot verify on the day you read it.
The design detail worth praising: the agent previews cost before spending. Video is the one medium where a sloppy prompt loop can burn a budget in minutes, and a mandatory cost preview is exactly the guardrail a team needs. Treat the free 100 credits as your evaluation budget — enough to test a full UGC-ad recipe end to end.
Who Popcraft is really for
Marketing teams and agencies. If your creative process already starts with a brief typed into a chat, Popcraft removes the context-switch to a separate studio. The recipe workflows — the 15-second vertical UGC ad flow, the thumbnail-with-identity-lock flow — are clearly built for people shipping ad creative and social content on deadline.
Developers building agent workflows. Because it is MCP-native, Popcraft drops into content pipelines, internal tools and automation scripts with one endpoint. If you are wiring AI video into a larger system — a content engine, a product-listing generator, an internal marketing tool — this is the integration surface you actually want.
Not for: editors who want a timeline. There is no manual fine-cutting here, and the maker is honest that canvas and character tools are still on the way. If your craft is frame-by-frame control, keep your editor and use Popcraft (or a raw API) only for raw clip generation.
Watch-outs and honest limitations
- Credits burn fast. At 42 credits per second for 720p Seedance, a handful of exploratory generations can eat the free tier. Always iterate at 480p and lean on the cost preview.
- Canvas and character tools are still coming — the maker’s own words. Recurring characters and complex scenes are improving across the industry, but verify identity consistency on your own briefs before promising it to a client.
- Human review is non-negotiable. AI video still produces wrong hands, garbled product details and off-brand moments. Nothing client-facing should ship without a human pass.
- Rights and privacy. Your briefs, product photos and reference images flow through a third-party service. Read the terms for commercial use and data handling before running client assets through it — especially regulated or confidential material.
- Model availability moves. The 20+ model roster is a snapshot; models get swapped as vendors ship. The cost preview and the recipes are the durable value; any specific model is not.
How Popcraft compares to the alternatives
The honest comparison set has three layers. First, raw model APIs (Replicate, fal.ai and the vendors’ own endpoints): potentially cheaper per generation, but you build the agent tooling, cost controls and workflow recipes yourself. That is a real engineering project, not a weekend hack.
Second, dashboard video studios — the direct Veo/Kling apps and AI video editors. Better for hand-crafted work where a human drives every decision, but they live outside your agent workflow. Every brief means exporting context, switching apps and re-uploading references.
Third, agent-driven video platforms like invideo, which routes shots to the right model automatically. This is Popcraft’s closest competition. Popcraft’s differentiators: it is MCP-native rather than a single app (so it works inside your agent, not theirs), it previews cost before spending, and it ships curated production recipes instead of raw prompts. If “works inside the tools my team already uses” matters to you, that is a meaningful edge.
From our experience: making it work in a real pipeline
We have spent the last year wiring AI tools into production workflows at MTD Technologies — content engines, automation pipelines, client marketing systems — and the pattern with tools like Popcraft is consistent. The teams that get value follow a discipline:
- Storyboard in text first. The recipe workflow exists for a reason — intake, identity, storyboard, realism pass, then generation. Teams that skip straight to generation pay for it in credits and rework.
- Iterate cheap, render once. Draft at 480p, lock the creative, then do the final 720p pass. With plan discounts the final render is where the rate cuts matter.
- Lock identity early. For recurring faces, products or brand elements, get the reference/identity-lock step right on the first generation. Fixing a drifting character across ten clips is the most expensive mistake in AI video.
- Keep a human gate on anything public. The agent can generate; only a person should approve. This is true of every AI tool we have reviewed, and video — where errors are visible — is the least forgiving medium of all.
- Track credit burn per project. Cost preview per generation is good; a weekly credit ledger per client is what keeps the economics honest.
Frequently asked questions
Does Popcraft work with Claude Code and ChatGPT?
Yes. Popcraft is an MCP server, so it works with anything that speaks MCP: Claude, Claude Code, Cursor, ChatGPT, Codex and similar hosts. You paste the MCP endpoint URL, sign in with OAuth, and the tools appear inside your agent.
How much does Popcraft cost?
You start with 100 free credits. Generation is billed in credits per second of output — for example, Seedance 2.5 lists at 42 credits per second at 720p, or 24 per second at 480p. Pro and Max plans discount that rate (up to 60% off on annual Max) and include a daily Seedance 2.0 Fast & Mini allowance. Check the live pricing page for current dollar prices.
Which AI models does Popcraft support?
Twenty-plus models across video, image, audio, 3D and upscaling, including Seedance 2.5, Veo 3.1, Kling 3.0 Omni, Google Omni 1.1, Nano Banana Pro, GPT Image 2.5 and ElevenLabs. The roster is updated as vendors ship new models.
Do I need an API key to use Popcraft?
No. You sign in with OAuth and the token auto-refreshes — there is no API key to generate, store or rotate. That is a genuine operational simplification for teams.
Is Popcraft worth it for a small business?
If your team already briefs work inside an AI agent and you need video or ad creative regularly, the free 100 credits are a no-risk way to test the full workflow. Occasional users making one video a month will likely do fine with a pay-per-generation dashboard instead. The value compounds with volume and with how deeply your workflow is already agent-driven.
Popcraft is a good example of where AI tooling is heading in 2026: the interface is disappearing into the agent you already talk to. For more tools in this space, browse our AI Tools Directory — and if you want AI video wired into your own content pipeline rather than bolted on beside it, talk to us. Building agent-integrated systems is what we do.