Claude doesn't have a built-in video generator — it's a text (and vision-input) model. What it does have is MCP (Model Context Protocol) support, which lets it call outside tools. Muapi's MCP server is one of those tools: connect it once, and Claude can generate video from text or animate an existing image throughmuapi_video_generateandmuapi_video_from_image, using the exact same API key you'd use for a direct API call. Setup and testing are free with a sandbox key; real generations are billed per second of output, from about $0.08/sec.
pip install muapi-cli (or npx -y muapi-cli without installing anything globally). It bundles the MCP server Claude will connect to.claude mcp add muapi -e MUAPI_API_KEY=your_key -- muapi mcp serveThen run claude mcp list and confirm it shows “muapi ✔ Connected”. Use the stdio transport shown above, not the hosted HTTP URL — that's the one setup gotcha.Every clip below is a real generation made by Claude through the connected MCP tool — same character carried across models, not disconnected stock footage.
The highest-quality text-to-video model in the catalog — used here for a cinematic establishing shot with camera motion, from a single prompt via muapi_video_generate.
The real “fire off a handful of variations” model — cheap enough per second to batch-generate several quick takes on one idea without thinking about cost.
Feed it a reference image and it keeps that identity locked while placing it in a brand-new scene and action — the payoff move for anyone building a recurring character.
Also behind the same key: Seedance 2.5 Image to Video (animate a still via muapi_video_from_image), uncensored video generation (Seedance 2.5 Spicy), and open-weights video (LTX-2 Pro) — see the full AI Video API catalog for every model and current ranking.
Sandbox testing is genuinely free and unlimited (mock output only). Real video is billed per second of rendered output, not per generation — so it costs more than the image side of this, and the price range is wider depending on model and resolution. No subscription, no seat minimum, no markup on top of calling the API directly.
| Model | Price | Best for |
|---|---|---|
| Seedance 2 Mini | $0.08 / sec at 480p | Cheap batch variants |
| Seedance 2.5 | $0.17–$1.70 / sec | Highest text-to-video quality |
| Seedance 2.5 Image to Video | $0.17–$1.70 / sec | Animating an existing still |
| Seedance 2.5 Omni Reference | $0.17–$1.70 / sec | Same character, new scene |
No. Claude is a text (and vision-input) model with no built-in video generator. Connecting an MCP tool like Muapi's is what gives a Claude conversation the ability to produce video.
Setting up and testing the connection is free — a sandbox API key returns an instant mock URL with no credit charge. Real generations are billed per second of rendered video, from $0.08/sec on the cheapest model and tier up to $1.70/sec on the flagship model at 4K, with no subscription or minimum spend.
Video is billed per second of output, not per generation, so the per-call price is inherently higher than a single image. A 5-second clip on Seedance 2.5 at 720p runs a little under $1.70 total — cheap for what it produces, but not free the way sandbox testing is.
All of Muapi's text-to-video and image-to-video models, through the same two MCP tools — muapi_video_generate and muapi_video_from_image. That includes Seedance 2/2.5, Kling, LTX-2, PixVerse, Sora 2, Runway, and MiniMax Hailuo. See the full ranked list on the AI Video API page.
Yes — muapi_video_from_image takes a reference image plus a motion prompt and returns an animated clip, so Claude can turn a still (one it or you already generated) into motion using the same connected tool.
Yes. Muapi exposes both a stdio MCP server (what Claude Code's claude mcp add uses) and a hosted HTTP MCP server for Claude Desktop, Cursor, Windsurf, and any other MCP-compatible client.