- Use Pydantic models and explicit operation_ids.
- No multipart; accept audio_base64, audio_url, or audio_path via JSON.
- Ensure OpenAPI spec is fully self-describing for MCP tool generation.
- Remove multipart form; use JSON with audio_base64 or audio_url.
- Add internal audio_path field for internal use.
- Keep OpenAPI spec consistent with new schema.
- New mcp_server.py: FastAPI app for LLMs to upload audio and get transcript JSON.
- Added process_mcp_transcribe_task Celery task.
- Updated __main__.py: WebUI always runs; MCP server runs in parallel when MCP_SERVER_ENABLED=true.