Tavus Video Replicas MCP. Generate talking-head videos from scripts and clones.

This server connects your agent directly to Tavus, letting you manage and generate personalized AI video content. You're running trained virtual clones—replicas—to create high-quality talking-head videos on demand.

Managing Your Replicas.
list_replicas gives you a complete rundown of every single trained video replica available to your agent. If you need specifics, get_replica_details pulls up all the data for one particular clone’s settings. You can also permanently delete unused clones using delete_replica, which cleans out that storage space.

Generating Content.
Need a new video? Just use generate_replica_video. You feed it a script and tell your agent exactly which replica to use; it handles the rest. When you’re ready, your agent can check the status of any job with get_video_details or get general details about the content using list_generated_videos. The system also lets you list every video asset that's been created on the platform via list_generated_videos.

Viewing and Cleaning Up.
To see everything you’ve built, your agent runs list_generated_videos, giving you a complete view of all generated videos along with their direct download links. If you only need to know about one specific video, use get_video_details for the granular info. When you're done with a piece or a clone, you can hit delete using delete_video to remove a finished video file, or delete_replica again if that trained replica isn't needed anymore.

This setup lets your agent manage every part of the process—from listing what clones you have (list_replicas) and checking their settings (get_replica_details), right through generating the talking-head video (generate_replica_video). It also tracks progress by providing detailed status checks on any given job, confirming when it's good to download. You can verify every piece of content with list_generated_videos, which returns a comprehensive list and associated links for all your finished work.

Cleaning up is simple: use delete_video if you want to wipe out a specific asset, or call delete_replica when a trained model is obsolete.

You don't have to juggle multiple systems to keep track of everything. Your agent keeps the content library clean by allowing you to list every existing video (list_generated_videos) and get granular details on any single piece (get_video_details). The workflow runs entirely through natural language commands, making it fast. If you change your mind about a specific replica's parameters or need to confirm its existence, get_replica_details gives you the definitive data.

This whole operation lets you treat your personalized video content like a managed asset vault. You get the list (list_replicas), check the specifics (get_replica_details), generate new stuff using those tools (generate_replica_video), track its readiness status, and then either keep it forever or delete it with delete_video or delete_replica. It's a closed loop of creation, monitoring, and cleanup.