---
name: fazita
description: Make videos, images and voice-overs with Fazita over its MCP server. Use when the user asks to generate a video, image, clip, or speech/voice-over, or mentions Fazita.
---

# Fazita

Fazita turns prompts into videos and images, and scripts into speech, through an MCP server at https://fazita.com/api/mcp.

## Set up (once)

If the `fazita` MCP tools aren't available:

1. Ask the user to open the MCP page in their Fazita studio (https://fazita.com/app/mcp) and make a key. Never ask them to paste the key into chat; have them run the command themselves.
2. They add the server:

```bash
claude mcp add --transport http fazita https://fazita.com/api/mcp --header "Authorization: Bearer YOUR_KEY"
```

3. Restart the session so the tools load. If tools answer "Your plan doesn't include MCP access", the plan needs upgrading: https://fazita.com/app/billing

## Making videos and images

1. Call `list_models` first, every time. It lists only the models the user's plan includes, with their ratios, qualities, lengths (`durations_sec`) and `credits` per take. Never assume a model, setting or price from memory: they change.
2. Call `get_usage` when credits matter. Tell the user what a take costs before queueing many, and don't queue more than they have left (`credits_left`).
3. Call `create_take` once per prompt with a model id from `list_models`. Pass `mode`, `ratio`, `quality` and `duration_sec` only from that model's listed options. For an image-to-video take, pass `reference_image_url`.
4. Takes run in the background. Poll `get_takes` every 10–20 seconds with the ids until each is `ready` or `failed`. Don't poll faster.
5. A ready take has a `file_url` (valid for a day). Download it into the user's project if they want the file, or share the link. Fazita deletes take files 15 days after they're made, so download anything worth keeping.
6. A failed take gives its credits back. Report the `error` and offer to retry or switch model.

## Speech

1. `list_models` shows voice models (`mode: "voice"`). Speech spends the model's characters, not credits (`characters_left`).
2. `list_voices` with the voice model id picks a voice.
3. `create_speech` answers with the audio when it's spoken; there is nothing to poll.

## Costs

Each video or image model has its own credits per take, and each plan has its own credits per day or billing cycle. For the full table, read https://fazita.com/docs/credits, or just trust `list_models` and `get_usage`.

## Rules

- One prompt per request; the server refuses batches.
- Prompts are limited in length; keep them short and concrete.
- Takes made here share the plan with the studio.