AI Video Check MCP Server
# AI Video Check — MCP server
Let Claude, Cursor or any MCP client check whether a video was generated by AI.
This server connects your AI assistant to [AI Video Check](https://www.aivideocheck.net), a detector for AI-generated video (Sora, Veo, Kling, MiniMax, HeyGen and others). Give it a link or a local file, and it returns the probability of AI generation with separate scores for the picture and the sound.
It is a thin layer over the public [AI Video Check API](https://www.aivideocheck.net/en/developers): the same key, the same credits and the same results.
## Tools
| Tool | What it does | Cost |
|---|---|---|
| `check_video(url)` | Checks the video at a public link (TikTok, Instagram, X, YouTube, Facebook or a direct file URL). Waits up to about a minute for the result. | 1 credit per started minute of video |
| `get_check(check_id)` | Returns the result of an earlier check. Results are kept for 24 hours. | free |
| `get_balance()` | Shows the remaining credits and the limits of your key. | free |
| `check_file(path)` | Checks a video file on your own computer. Local mode only. | 1 credit per started minute of video |
Every result tells the assistant how to read it: a probability, not proof. A low score means no signs of generation were found, not that the video is authentic. Editing, visual effects and face swaps on real footage are not detected.
## Get a key
1. Sign in at [aivideocheck.net](https://www.aivideocheck.net) and confirm your e-mail.
2. Open **Dashboard → API** and create a key. It is shown once.
3. Your first key comes with 20 free credits for the API.
Keep the key secret. Anyone holding it spends your credits. You can revoke a key on the same page; it stops working within a minute.
## Connect: remote server (links)
The server runs at `https://api.aivideocheck.net/mcp` (Streamable HTTP). Your client sends the key in the `Authorization` header.
**Claude Code**
```bash
claude mcp add --transport http aivideocheck https://api.aivideocheck.net/mcp --header "Authorization: Bearer avc_live_..."
```
**Cursor** (`~/.cursor/mcp.json`)
```json
{
"mcpServers": {
"aivideocheck": {
"url": "https://api.aivideocheck.net/mcp",
"headers": { "Authorization": "Bearer avc_live_..." }
}
}
}
```
Any other client that supports Streamable HTTP with custom headers works the same way.
## Connect: local server (links and files on your computer)
The local server runs on your machine and can also check video files from your disk. It uploads them to AI Video Check for analysis, exactly as the website does; files are deleted when the check ends.
```bash
git clone https://github.com/aivideocheck/aivideocheck-mcp
cd aivideocheck-mcp
python -m venv .venv
.venv/bin/pip install -r requirements.txt # Windows: .venv\Scripts\pip
```
Then point your client at it, for example **Claude Desktop** (`claude_desktop_config.json`):
```json
{
"mcpServers": {
"aivideocheck": {
"command": "/path/to/aivideocheck-mcp/.venv/bin/python",
"args": ["/path/to/aivideocheck-mcp/avc_mcp.py"],
"env": { "AVC_API_KEY": "avc_live_..." }
}
}
}
```
`check_file` sends only files that are named and built like a video container (MP4, MOV, WebM, MKV, AVI and similar). Anything else stays on your computer, so an instruction hidden in a web page cannot make your assistant upload, say, your SSH key.
## Configuration
| Variable | Default | Meaning |
|---|---|---|
| `AVC_API_KEY` | — | Your key. Read in local mode only; the remote server never uses a key from its own environment. |
| `AVC_API_BASE` | `https://api.aivideocheck.net` | API address. |
| `AVC_MCP_WAIT_S` | `55` | How long `check_video` waits for a result before returning the check id. |
## Development
```bash
.venv/bin/python tests/test_mcp.py
```
The test runs the server against a fake API on 127.0.0.1, over real HTTP and real stdio, with the MCP SDK's own client. It does not touch the live service.
## Pricing and terms
Checks spend the same credits as the website: one credit per started minute of video. Prices are on the [developer page](https://www.aivideocheck.net/en/developers#pricing). Use of the API is governed by the [Terms of Service](https://www.aivideocheck.net/en/legal/terms) and the [Privacy Policy](https://www.aivideocheck.net/en/legal/privacy).
Questions: support@aivideocheck.net
## License
MIT — see [LICENSE](LICENSE).
TDQS
Scored across 4 tools
check_video and check_file both detect AI-generated video and differ only by input source (public link vs local file), which the descriptions make clear but still requires reading carefully. get_check and get_balance have clearly distinct roles (retrieve result vs account status), so overlap is limited to the two check tools.
All four names follow a strict verb_noun snake_case pattern (check_video, check_file, get_check, get_balance). The verbs are used coherently: check_ for initiating analysis, get_ for retrieving information.
Four tools is well-scoped for a focused detection service: two submission paths, one result retrieval, and one account query. Every tool earns its place with no redundancy or padding.
The submit/retrieve/balance lifecycle is covered, including both URL and file inputs. Minor gaps exist: no way to list or cancel prior checks, and results expire after 24 hours without a cleanup or history operation.