AI Image, Video, and Music Creation

Problem

When an agent creates images, videos, music, or speech, the hard part is not the model call but the billing, async task, and asset-reference discipline. Remote MCP endpoints cannot read local files directly, duplicate submissions can create extra charges, and a queued task can be mistaken for a failure. @beatra-ai/beatra-ai turns Beatra media creation into an explicit execution contract, letting an agent complete tasks through the smallest matching workflow.

How It Works

• Fixed transport: Beatra tools run through the bundled scripts/mcp_client.py against an authenticated Beatra MCP endpoint, rather than relying on a host Beatra Connector.
• Idempotent generation: each logical generation gets one client_request_id; if the first response is lost, retry with the same ID, but use a new ID when any input changes.
• Async task handling: persist the task_id and poll beatra.tasks.get until terminal; do not create another paid task while one is queued or running.
• Local uploads: local image, audio, or video files are uploaded via beatra.assets.upload, completing the HTTP PUT before using the returned artifact ID.
• Verifiable results: report task status, artifact links or IDs, resolved model, usage, and net charged credits only when the response provides them.

The skill focuses on media creation and management, not on replacing the host agent's native text, image, or video understanding. Model selection, billing recovery, and uninstall/disconnect should be handled by the matching workflow to avoid breaking the shared Beatra connection.