Vidrush MCP

Connect an MCP client to Vidrush for video, image, music, and audio generation, task polling, and credits.

Endpoint and authentication

Vidrush exposes a remote Streamable HTTP MCP server:

https://vidrush-ai.com/api/mcp

Two authentication methods are supported. The server name is vidrush.

OAuth (claude.ai and Claude Desktop). Add a custom connector named Vidrush with the URL above, click Connect, and sign in with your Vidrush account. Clients discover the flow automatically through /.well-known/oauth-protected-resource; no API key is needed.

API key (Claude Code, Cursor, Codex, SDKs). Configure an MCP client that supports HTTP headers to send:

Authorization: Bearer sk-your-api-key

A request without a valid OAuth token or API key is unauthorized.

Tools

The Vidrush MCP server generates video, image, music, and audio. Generation consumes credits — an agent should inspect a model and tell the user about the cost before it calls a generate_* tool.

Shared tools

Tool	Input	Result
`get_task`	`task_id`	Current `status`, `media_type`, and when available `output_urls` (for video also `video_urls`)
`wait_for_task`	`task_id`, optional `timeout_seconds`	Polls and returns the latest task; default 45 seconds, minimum 1, maximum 50
`get_credits`	none	`remaining_credits` for the authenticated account

Video

Tool	Input	Result
`list_video_models`	none	Video model IDs, labels, vendors, modes, durations, resolutions, credit labels
`get_video_model`	`model`	Full capability and pricing data for one video model
`generate_video`	`model`, `prompt`, optional `mode`, `image_urls`, `video_urls`, `duration`, `resolution`, `aspect_ratio`, `audio`, `source_video_duration_seconds`	Creates a task; returns `task_id`, `status`, `cost_credits`, polling hint

Image

Tool	Input	Result
`list_image_models`	none	Image model IDs, modes, aspect ratios, qualities, credit labels
`get_image_model`	`model`	Full capability and pricing data for one image model
`generate_image`	`model`, `prompt`, optional `scene`, `image_urls`, `aspect_ratio`, `quality`	Creates a task; poll `get_task` for `output_urls`

Music

Tool	Input	Result
`list_music_models`	none	Music models, controls, and credit cost
`generate_music`	`prompt`, optional `duration_seconds`, `instrumental`, `style`, `lyrics`	Renders synchronously; usually returns `output_urls` immediately

Audio

Tool	Input	Result
`list_voices`	none	Text-to-speech voices (`id`, `name`, `gender`, `accent`, `language`) and credit cost
`generate_speech`	`text`, `voice_id`, optional `model_id`, `speed`, `stability`, `similarity_boost`	Renders synchronously; usually returns `output_urls` immediately
`generate_sound_effect`	`prompt`, optional `duration_seconds`, `prompt_influence`	Renders synchronously; usually returns `output_urls` immediately

The same model validation and credit rules as the v1 API apply.

Typical workflow

Call the matching list_* tool (for example list_video_models or list_voices), then get_*_model for the selected model.
Confirm that generation will consume the displayed credits.
Call the matching generate_* tool.
Call wait_for_task, or call get_task every 5 seconds until the status is success, failed, or canceled. Music and audio usually complete immediately.
On success, return output_urls to the user.

Errors and recovery

Tool failures use MCP error results with an actionable text message. Insufficient-credit errors include the Vidrush pricing URL. For an invalid model or option, call get_video_model again and correct the input instead of retrying the billable tool blindly.

Vidrush MCP

On this page