Video feedback lives or dies on being able to point at the right moment. Fetching a frame by timestamp is expected; fetching it by a spoken word means an agent can resolve a vague human reference into an exact frame without anyone scrubbing a timeline. wait_for_feedback is the other unusual one — a blocking call that returns when a colleague comments, which lets an agent wait on a human instead of polling. And there is no delete tool anywhere, which is the right absence for a record of what people said.
Flask's endpoint for video feedback collaboration, which listed 17 tools to our probe as flask 1.0.0. It reads and writes comments but cannot delete them, and it can block waiting for new feedback to arrive.
- contents, search, recent_activity, tags — browse the team's folders and assets, search across folders, assets and feedback at once
- feedback_list, feedback_get, feedback_post, feedback_update — comments with tags and video timestamps, including reply threads
- wait_for_feedback — block for up to 45 seconds and return the moment new feedback lands
- get_annotated_frames and get_frame — the visuals behind a comment, or the exact frame at a given time or word
- asset_status — processing progress for a video still being encoded
- permission_get — who can reach a folder and at what level
- upload_video from a public URL, or upload_file_start and upload_file_complete for a local file
- feedback_stats — counts per asset and per version across many assets in one call
OAuth 2.0 with dynamic client registration. Streamable HTTP, POST only — no SSE. The vendor states it accesses only what a tool call asks for, with no background sync.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen