# Sparkiz MCP server: create AI video ads from any MCP client

> This document is the machine-readable twin of https://sparkiz.ai/connect-ai
> (Hebrew: https://sparkiz.ai/he/connect-ai). It is written for AI agents and
> developers. Last updated: 2026-07-17.

## What Sparkiz is

Sparkiz (https://sparkiz.ai) is an AI video ad generator. It turns a product
link or a short brief into a finished, ready-to-post MP4 video ad with an AI
actor, script, voiceover, captions, and music. Videos are produced in Hebrew,
English, or Russian, in 9:16, 1:1, or 16:9, with full commercial rights.

Sparkiz exposes this capability over MCP (Model Context Protocol), so an AI
assistant such as Claude, ChatGPT, or any MCP-capable client can create video
ads inside a conversation.

## Availability

Early access, rolling out now. General availability arrives with OAuth
sign-in. Users get their personal connect link in the Sparkiz app under
Settings > Connect AI (the tab appears as access reaches their account).
This document is kept current.

## How to connect

1. The user copies their personal connect link from the Sparkiz app:
   Settings > Connect AI > Create link. The link is account-scoped and
   revocable at any time (up to 3 live links per account).
2. Add it to the client:
   - Claude (web/desktop): Settings > Connectors > Add custom connector > paste the link.
   - ChatGPT: enable Developer mode, then Settings > Connectors > add by URL.
   - Claude Code or another CLI: `claude mcp add --transport http sparkiz <connect link>`
   - Any other MCP client: add an MCP server by URL with the connect link.
3. Ask for an ad. Example: "Make a 15 second video ad for this product page: <url>".

Transport: Streamable HTTP, stateless. The server self-documents: call
`sparkiz_get_started` first; every tool description states when to use it.

## Canonical workflow for agents

```
sparkiz_get_started
  -> sparkiz_get_preferences        (ask the user for any unknown preference:
                                     language, aspect ratio, video length, avatar)
  -> sparkiz_quote_ad               (0 credits, returns the exact price)
  -> USER CONFIRMATION              (required; credits are real money, never skip)
  -> sparkiz_create_ad
  -> sparkiz_get_status             (poll; surfaces checkpoints and the video URL)
  -> sparkiz_approve_step           (answer checkpoints: images / voice / paid extras)
```

Competitor-ad cloning: `sparkiz_analyze_ad` (2 credits, produces a structural
blueprint) then `sparkiz_clone_ad` (quoted like create_ad). Cloning copies
structure only: scene count, pacing, caption style, presenter framing. It
never copies words, frames, or assets.

## Tools (15)

| Tool | Purpose | Cost |
|---|---|---|
| sparkiz_get_started | Learn the product and the workflow; call first | 0 credits |
| sparkiz_get_preferences | Read saved defaults (language, aspect, length, avatar, autonomy) | 0 credits |
| sparkiz_save_preferences | Save those defaults for future runs | 0 credits |
| sparkiz_list_avatars | AI actors matched and scored against the ad's casting profile | 0 credits |
| sparkiz_quote_ad | Exact credit price for a described ad, before building | 0 credits |
| sparkiz_create_ad | Build the ad end to end | Quoted, user-confirmed |
| sparkiz_analyze_ad | Competitor ad to structural blueprint | 2 credits |
| sparkiz_clone_ad | Rebuild an analyzed structure with the user's product | Quoted, user-confirmed |
| sparkiz_get_status | Poll a job: step, checkpoints, result URL | 0 credits |
| sparkiz_approve_step | Answer a checkpoint (approve / adjust in app / skip) | 0 credits |
| sparkiz_cancel_job | Cancel a running build | 0 credits |
| sparkiz_list_videos | The user's recent videos | 0 credits |
| sparkiz_request_asset_upload | Private, time-limited, no-login upload page for user media | 0 credits |
| search | Read-only workspace search (standard connector format) | 0 credits |
| fetch | Read-only fetch of one item found by search | 0 credits |

## Money rules (enforced server-side)

- Every paid step requires a fresh quote plus explicit user confirmation.
- Credits are deducted when a step starts and refunded automatically on failure.
- Single attempt per paid call: no silent retries.
- A daily spend cap applies per account.
- Credits come from the user's regular Sparkiz plan (https://sparkiz.ai/pricing).
  There is no separate MCP pricing.

## Rules for agents

- Never skip a quote confirmation. Credits are real money.
- Never invent product facts, prices, or reviews. Ask Sparkiz or the user.
- Ask the user for unknown preferences (language, aspect ratio, video length)
  before building, in every autonomy mode.
- User uploads go through sparkiz_request_asset_upload; files are validated
  against what the scene needs, and the user always gets a use-anyway choice.
- Connect links are personal. Never log, echo, or share them.

## Links

- Human docs (English): https://sparkiz.ai/connect-ai
- Human docs (Hebrew): https://sparkiz.ai/he/connect-ai
- Product: https://app.sparkiz.ai/signup
- Pricing: https://sparkiz.ai/pricing
- Contact: info@sparkiz.ai