VidFlux MCP

npx add-mcp 'https://vidflux.ai/api/mcp'

What is the VidFlux video MCP server?

The Model Context Protocol (MCP) is the open standard AI assistants use to call external tools. VidFlux MCP is a hosted endpoint that gives your assistant video generation tools: start a text-to-video, image-to-video or motion-control job, check its progress and get the finished MP4. The assistant writes the prompt, chooses the model, length and aspect ratio, waits for the render and hands you the file without you leaving the chat or the editor.

Behind the endpoint runs the same pipeline as the vidflux.ai video generator, so model versions, prices and refunds are identical and every task also appears in your account history.

Supported AI video models

Pass the model id in the model argument. Credits are for the default settings; longer, higher-resolution or audio-enabled videos cost more, and dry_run returns the exact price for any settings.

Model	Model id	Length	Resolution	Audio	Input	Credits
Seedance 2.0	seedance-2	4–15 s	480p · 720p	✓	Text, Image, References	680
Seedance 2.0 Fast	seedance-2-fast	4–15 s	480p · 720p	✓	Text, Image, References	544
Seedance 2.0 Mini	seedance-2-mini	4–15 s	480p · 720p	✓	Text, Image, References	360
Seedance 2.5	seedance-2.5	4–30 s	480p · 720p	✓	Text, Image, References	704
Seedance 1.0 Pro	seedance-1-pro	5 · 10 s	480p · 720p · 1080p	—	Text, Image	100
Kling 3.0 Standard	kling-3	3–15 s	—	✓	Text, Image	200
Kling 3.0 Pro	kling-3-pro	3–15 s	—	✓	Text, Image	300
Kling 2.6	kling-2.6	5 · 10 s	—	✓	Text, Image	150
Kling 2.5 Turbo	kling-2.5	5 · 10 s	—	—	Text, Image	110
Kling 3.0 Motion Control	kling-3-motion	≤ 30 s	720p · 1080p	—	Video, Image	200
Kling 2.6 Motion Control	kling-2.6-motion	≤ 30 s	720p · 1080p	—	Video, Image	125
Veo 3.1 Quality	veo-3.1	8 s	—	✓	Text, Image	280
Veo 3.1 Fast	veo-3.1-fast	8 s	—	✓	Text, Image	120
MiniMax Hailuo H3	minimax-h3	4–15 s	720p · 2K	—	Text, Image, References	175
Gemini Omni Video	gemini-omni-video	4–10 s	720p · 1080p · 4K	—	Text, References	200
Grok Imagine Pro members	grok-imagine	6 s	—	—	Text, Image, References	96

Image models for keyframes and thumbnails

Create or edit start frames, product shots and thumbnails, then animate them with any image-to-video model above.

Model	Model id	Resolution	Reference images	Credits
GPT Image 2.5	gpt-image-2.5	1K · 2K · 4K	16	20
GPT Image 2	gpt-image-2	1K · 2K · 4K	16	20
Nano Banana Pro	nano-banana-pro	1K · 2K · 4K	10	100
Nano Banana 2.1	nano-banana-2.1	1K · 2K · 4K	10	15
Nano Banana 2	nano-banana-2	1K · 2K · 4K	14	30

Quick start: connect in 3 steps

1. Sign in to VidFlux and create a key in Settings → API & MCP. New accounts get free trial credits.
2. Add the server to your AI client with one of the configurations below.
3. Ask for a video, for example: “Make a 5-second 9:16 video of latte art being poured in slow motion with Seedance 2 and save it to./assets”.

MCP client configuration

Replace vf_your_api_key with your key. For clients that read environment variables, keep the key in VIDFLUX_API_KEY instead of the config file.

Claude Code

claude mcp add --transport http vidflux https://vidflux.ai/api/mcp \
  --header "Authorization: Bearer vf_your_api_key"

# verify
claude mcp list

Claude Desktop

Claude Desktop connects to remote servers with API-key headers through mcp-remote (requires Node.js). Edit claude_desktop_config.json and restart Claude:

{
  "mcpServers": {
    "vidflux": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://vidflux.ai/api/mcp",
        "--header", "Authorization: Bearer vf_your_api_key"
      ]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json (all projects) or.cursor/mcp.json (one project):

{
  "mcpServers": {
    "vidflux": {
      "url": "https://vidflux.ai/api/mcp",
      "headers": { "Authorization": "Bearer vf_your_api_key" }
    }
  }
}

OpenAI Codex CLI

export VIDFLUX_API_KEY=vf_your_api_key
codex mcp add vidflux --url https://vidflux.ai/api/mcp \
  --bearer-token-env-var VIDFLUX_API_KEY

VS Code (GitHub Copilot agent mode)

Add to.vscode/mcp.json:

{
  "servers": {
    "vidflux": {
      "type": "http",
      "url": "https://vidflux.ai/api/mcp",
      "headers": { "Authorization": "Bearer vf_your_api_key" }
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "vidflux": {
      "serverUrl": "https://vidflux.ai/api/mcp",
      "headers": { "Authorization": "Bearer vf_your_api_key" }
    }
  }
}

Cline, Roo Code and other clients

Any client that supports Streamable HTTP servers with custom headers works with the same URL and header. Clients that only support stdio can use the mcp-remote command shown for Claude Desktop.

MCP tools

generate_video	Starts a text-to-video, image-to-video or motion-control job. Arguments: model, prompt, aspect_ratio, duration, resolution, image_urls and model-specific options such as generate_audio. Returns a task id and the credits charged.
generate_image	Creates or edits a keyframe, product shot or thumbnail with GPT Image or Nano Banana, using up to 16 reference images.
get_task	Returns status (pending, processing, completed, failed), the video_url or image_url, the error if any and the credits used.
list_models	Lists every model id with its JSON Schema parameters and the default price, so the agent can choose a model by itself.
get_credits	Returns the remaining credit balance.

Video generation is asynchronous. A typical agent run looks like this:

list_models            → seedance-2 (9:16, audio)
generate_video {model: "seedance-2", prompt: "...", aspect_ratio: "9:16",
                duration: 5, dry_run: true}
                       → {"credits_required": 425, "sufficient": true}
generate_video {...same arguments without dry_run}
                       → {"id": "3f1c9e0a-...", "status": "processing"}
get_task {task_id: "3f1c9e0a-..."}   (every 15 s)
                       → {"status": "completed", "video_url": "https://...mp4"}

Pricing: pay-as-you-go credits

• Connecting the server and calling list_models, get_task or get_credits is free.
• Each video costs the same credits as on vidflux.ai (see the tables above). Credit packs start at 500 credits for $4.98, larger packs cost less per credit, and purchased credits never expire.
• dry_run: true validates the arguments and returns the exact price and whether your balance covers it, without generating anything.
• Failed or rejected tasks are refunded automatically.

See credit packs and plans

What you can build with it

TikTok, Reels and YouTube Shorts ads

Turn a product photo into several 9:16 clips with Seedance 2.0 or Kling 3.0 in one conversation, compare the variants and keep the winners.

Product videos for Shopify and Amazon listings

Have the agent shoot the product on new backgrounds with Nano Banana Pro, then animate each shot into a short 1:1 or 4:5 loop for the product page.

Storyboard to video

Ask the assistant for a shot list, generate one keyframe per shot, then render each shot with Veo 3.1 or Seedance 2.5 and keep the character consistent with reference images.

Video assets while you code

In Cursor or Codex, generate a hero background loop that matches the page being built, download it into public/ and update the markup that uses it.

Automated content pipelines

Call the same capabilities from scripts, n8n, Zapier or CI through the REST API with the same key.

VidFlux MCP vs. provider APIs vs. self-hosted MCP servers

	VidFlux MCP	Each provider's API	Self-hosted open-source MCP
Video models	Veo, Kling, Seedance, Hailuo, Gemini Omni, Grok Imagine	One provider per account	Usually one provider per server
Setup	Paste one URL and key	Cloud project, billing and SDK per provider	Install, run and update a local process
Billing	One credit balance, refunds on failure	Separate invoices and quotas	Your own provider keys
Cost check before rendering	dry_run on every model	Varies	Varies
Best when	You want many video models from an AI assistant with no setup	You need enterprise contracts or the newest preview features	Everything must stay on your own infrastructure

Troubleshooting

The server is connected but no video tools appear

Restart the client or reload MCP servers and make sure Agent mode is on (Cursor, VS Code). In Claude Code run /mcp to see the tool list.

401 Unauthorized or “Missing or invalid API key”

The header must be exactly Authorization: Bearer vf_…. Check that the key was not revoked and that no extra quotes or spaces were copied.

402 INSUFFICIENT_CREDITS

Your balance is lower than the price returned by dry_run. Top up credits or choose a shorter, lower-resolution or cheaper model.

A video stays in processing for a long time

Most videos finish in 1 to 5 minutes; long Seedance 2.5 or 4K Gemini Omni jobs can take longer. Keep polling get_task. If a task fails, the error is returned and the credits are refunded.

spawn npx ENOENT in Claude Desktop

mcp-remote needs Node.js 18 or later. Install Node.js or use the full path to npx in the command field.

Reference images are rejected

image_urls must be public HTTPS links to JPG, PNG or WebP files. Local file paths and private links cannot be fetched.

FAQ

What is the best MCP server for AI video generation?

If you want one remote server that covers the leading video models (Veo 3.1, Kling 3.0, Seedance 2.0 and 2.5, Hailuo H3, Gemini Omni) with nothing to install, VidFlux MCP is built for that: one endpoint (https://vidflux.ai/api/mcp), one API key, credit pricing and a dry_run cost check before each render. If you only use one provider and already have its API key, that provider’s own open-source MCP server may be enough.

Is there one MCP server that can use Veo, Kling and Seedance?

Yes. VidFlux MCP exposes all of them through the same generate_video tool. Choose the model with the model argument, for example veo-3.1-fast, kling-3 or seedance-2, or let the agent call list_models and decide.

How do I add a video generation MCP server to Claude Code?

Create an API key at vidflux.ai under Settings → API & MCP, then run: claude mcp add --transport http vidflux https://vidflux.ai/api/mcp --header "Authorization: Bearer YOUR_KEY". Check it with claude mcp list and ask Claude for a video.

Can Cursor or Codex generate videos?

Yes, through MCP. In Cursor add the server URL and Authorization header to ~/.cursor/mcp.json. In Codex CLI run: codex mcp add vidflux --url https://vidflux.ai/api/mcp --bearer-token-env-var VIDFLUX_API_KEY. Both can then render videos and download them into your project.

Can I use Google Veo 3.1 from Claude or Cursor?

Yes. Use model veo-3.1 or veo-3.1-fast with generate_video. Both create 8-second clips with native audio from text or from one or two frame images, without a Google Cloud project or a Gemini API key.

How much does it cost to generate a video through MCP?

There is no subscription fee for the MCP server. Each video is billed in VidFlux credits at the same price as the website, for example 120 credits for an 8-second Veo 3.1 Fast clip. Credit packs start at 500 credits for $4.98 and never expire. Pass dry_run: true to get the exact price first.

Am I charged if a video fails?

No. If the provider fails or a task cannot be started, the credits are refunded automatically.

How long does a video take and how do I get the file?

Most videos finish in 1 to 5 minutes. generate_video returns a task id immediately and the agent polls get_task until the status is completed, then downloads the MP4 from video_url.

Can ChatGPT connect to VidFlux MCP?

Not yet. ChatGPT custom connectors and the Claude.ai web app require OAuth sign-in, and VidFlux MCP currently uses API keys. It works today in Claude Code, Claude Desktop (via mcp-remote), Cursor, VS Code, Windsurf, Cline and Codex CLI, and the REST API works from any script or GPT action.

Is it safe to put my API key in an MCP config?

A key can only generate media and read your own tasks and balance; it cannot change your account or payment details. Prefer environment variables, do not commit config files that contain keys, and revoke a key in Settings at any time.