Visiojoy MCP docs

Connect

1. Add a custom MCP connector in Claude, Cursor, or your MCP client of choice, using the URL below as the server address.
2. You'll be redirected to sign in and authorize — no API key needed.
3. Once connected, your assistant gets seven tools: four that create (generate_image, generate_video, remove_background, upscale_image) and three free, read-only ones (get_task_result, list_models, get_credit_balance).

claude mcp add --transport http visiojoy https://mcp.visiojoy.com

{
  "mcpServers": {
    "visiojoy": {
      "url": "https://mcp.visiojoy.com/mcp"
    }
  }
}

Examples

You normally just ask your assistant in plain language and it fills in the tool call. These are the calls it sends, with the credits each one costs today.

{
  "prompt": "A ceramic coffee mug on a sunlit wooden table, soft morning light, product photo",
  "model": "nano-banana-2-1",
  "size": "16:9",
  "quality": "2K"
}

{
  "prompt": "Place the product on a white marble kitchen counter, keep the label unchanged",
  "images": [
    "https://example.com/product.jpg"
  ],
  "model": "nano-banana-2-1-edit",
  "quality": "1K"
}

{
  "image": "https://example.com/portrait.png"
}

{
  "prompt": "Slow waves rolling onto a sandy beach at sunset, gentle camera push-in",
  "model": "bytedance/seedance-2-fast",
  "resolution": "720p",
  "duration": 5,
  "aspectRatio": "16:9"
}

{
  "prompt": "The product rotates slowly on a turntable, soft studio lighting",
  "images": [
    "https://example.com/product.jpg"
  ],
  "model": "wan/3-0-video",
  "resolution": "480p",
  "duration": 5
}

Generation complete (text-to-image / nano-banana-2-1), 1 image(s) total.
resource_link  image-1  https://…/generations/…/0.png

Task is still processing (taskId: …). You can use the get_task_result tool with this taskId to check again.

Video task submitted (taskId: …). Seedance 2.0 Fast, text-to-video, 720p, 5s — 125 credits reserved, charged only if the video is delivered. Videos usually take a few minutes: call get_task_result with this taskId to fetch it.

Generation complete (text-to-video / bytedance/seedance-2-fast), 1 video(s) total.
resource_link  video-1  https://…/generations/…/0.mp4

Tools

Models & credits

Pass the model ID as model. For images, pass one of its listed values as quality; prices are per image. Videos are priced per second. All prices match the Visiojoy website. New accounts get 12 free credits; see pricing for plans and credit packs. Your assistant can fetch the same data with list_models.

• All accounts — available on the free tier.
• After any purchase — unlocked by any subscription or credit pack.
• Active subscription — subscription plans only.
• Video models are available to all accounts.

Image tools

Video costs

A video costs seconds × the per-second rate of the chosen resolution. That adds up quickly: Seedance 2.5 at 1080p for 30 seconds costs 4,740 credits. Because an assistant can call tools on its own, generate_video is built to avoid surprise spending:

• Cheapest settings by default. Without a resolution it uses the model's cheapest one, and the model's default length. Every model's default call costs 18–140 credits (see Default call in the table above).
• A 300-credit cap per call. If a request would cost more, nothing is submitted or charged, and the error states the exact cost. The assistant is told to confirm with you before calling again with a higher maxCredits.
• Charged only on delivery. The cost is reserved when the task is submitted and returned automatically if the video fails.
• The cost is always reported. The submit message states how many credits were reserved.

How results come back

generate_image waits about 16 seconds for the result before returning. There are two outcomes:

Credits

• The cost is reserved when the task is submitted.
• It is only deducted when the image or video is delivered.
• If the task fails, the reserved credits are returned automatically.

Files, watermark and license

• Images and videos are stored on Visiojoy and show up in your Assets alongside website generations.
• Same rules as the website: results from accounts that have never made a purchase can be used for personal, non-commercial purposes only, and images may carry a small watermark (videos are not watermarked).
• After your first subscription or credit pack, new results are watermark-free and licensed for commercial use. See the Terms of Service.

Errors

Errors come back as a tool result with isError: true and a readable message, so your assistant can explain the problem or retry with different settings.

Limits

• One image or one video per call. Ask again for more variations.
• remove_background and upscale_image take one image per call.
• Videos support text-to-video and image-to-video. Reference videos and audio (the reference mode on the website) are not available through MCP yet.
• Up to 4 input images per image call and 7 per video call, fewer for most models (see the tables above). Input URLs must be publicly reachable.
• Results are private unless you set isPublic to true.