Use these requests after you connect a compatible agent and prepare visual search when a prompt asks about scenes or subjects. Replace the example filenames and bracketed details with your own.
The retrieval examples use the local, read-only tools described in the MCP overview. For the reel prompt, describe your product and the reaction you want; the agent finds matching clips through MCP and creates a timed edit plan. It can also inspect detailed ranges and save a native shotlist when those tools are available. Text overlays remain editing directions; rendering the final video requires an editing tool.
Find and inspect a described shot
- Starting point
- Visual search is prepared and the MCP server is running.
- Goal
- Find the strongest sampled match and return a link to the moment.
- Tools
library_stats,search_library,get_contact_sheet,get_frame,get_video_at_timestamp
Prompt
Find footage matching this description: a person walking beside the sea. Check library_stats, then search_library in visual mode. Inspect the strongest results with get_contact_sheet or get_frame. Return clip IDs, filenames, actual frame timestamps and clickable open URLs. Explain that matches are based on sampled frames and that similarity scores are not confidence probabilities.Based on the built-in find_footage prompt. Replace the scene description with the shot you need.
Show metadata for a clip
- Starting point
- You know part of the clip filename. Replace 16928387 with your own.
- Goal
- Return the matching clip metadata and available sampled frames.
- Tools
search_library,get_clip_metadata
Prompt
Show metadata for a clip named 16928387. Include the filename, duration, dimensions, relative source path, and the available sampled frame timestamps.Find short landscape clips
- Starting point
- The library has been prepared and you want a quick shortlist.
- Goal
- Find landscape clips under ten seconds and compare their contact sheets.
- Tools
search_library,get_contact_sheet
Prompt
Find landscape clips under ten seconds, and show their contact sheets. Include each clip ID, filename, duration, and the sampled timestamps represented in the sheet.Inspect the best visual match
- Starting point
- Visual search is prepared for the connected library.
- Goal
- Find a scene, inspect the strongest frame, and provide a timestamp link.
- Tools
search_library,get_frame,get_video_at_timestamp
Prompt
Find a park with trees and city buildings. Show the best matching frame, its actual timestamp, the clip filename, and a clickable link that opens the native app at that moment.Ask the agent to verify the frame before choosing it; visual ranking is not a confidence score.
Find a subject in a setting
- Starting point
- Visual search is prepared and the agent can inspect cached frames.
- Goal
- Find a person by the sea and open the most useful moment.
- Tools
search_library,get_contact_sheet,get_frame,get_video_at_timestamp
Prompt
Find a person by the sea. Inspect the strongest match and give me the clip filename, the actual frame timestamp, and a link to open that moment in B-Roll Archives.Build a filtered shortlist
- Starting point
- You want to narrow the local catalog before reviewing images.
- Goal
- Apply metadata filters, then inspect only the most relevant ready clips.
- Tools
library_stats,search_library,get_contact_sheet,get_frame
Prompt
Search the local library for ready landscape clips between five and ten seconds long. Return the strongest results with their IDs, filenames, durations, and contact sheets. Then recommend which sampled frame I should inspect first.The server can filter by readiness, orientation, and duration without exposing arbitrary filesystem paths.
Build a 30-second product reel
- Starting point
- Your library contains product demonstrations and reaction footage. Describe the product, desired reaction, benefit, hook, overlays, and CTA below; the agent will find the clips through MCP.
- Goal
- Pair the most intriguing demo moment with a hook and a three-second reaction, then return a complete 30-second edit plan.
- Tools
library_stats,search_library,get_clip_metadata,get_contact_sheet,get_frame,get_video_at_timestamp- When available
inspect_clip_range,get_library_job,create_shotlist
Prompt
Build a 30-second product reel using footage from my library.
Product/demo: [describe the product and the action or feature to show]
Main benefit: [what the viewer should understand]
Hook: [opening idea, or suggest one]
Reaction: [describe the reaction you want, e.g. happy and excited]
Text overlays: [exact wording, or suggest concise copy]
Closing CTA: [desired action]
Check library_stats, then use search_library in visual mode to find the product demo, suitable hook footage, and the requested reaction from these descriptions. Inspect the returned images before choosing clips. Focus the demo on its most intriguing action or reveal and match the reaction to that moment. Use exactly three continuous seconds of the reaction around its clearest expression of the requested emotion. Verify the expression from the footage.
Start with this timing: hook 0–3s, demo 3–24s, reaction 24–27s, closing CTA 27–30s. You can rearrange the beats if it improves the story, but keep the total exactly 30 seconds and the reaction exactly three seconds. Default to vertical 9:16 and flag footage that needs a crop.
Check library availability, clip durations, and candidate frames before selecting ranges. If inspect_clip_range is available, use it to refine the demo and reaction moments and poll get_library_job for the results. Distinguish inspected samples from proposed cut points, and keep every range within its source clip. If a clip is missing, too short, or does not match, explain the gap instead of inventing footage.
Return a timeline with clip IDs, filenames, source in/out times, reel timings, overlay text and timing, crop directions, and links to the selected moments. The durations must sum to exactly 30 seconds, including the closing CTA.
If create_shotlist is available, save the ordered sequence in B-Roll Archives using current clip revisions and one unique idempotency key for this sequence. Put overlay text, timing, and crop directions in each shot's notes, and return the shotlist link. Otherwise, return the complete edit plan.Range inspection and saved shotlists depend on the tools exposed by your connected app. Overlays are editing directions stored in notes; rendering a video and applying the overlays requires an editing tool.
The server returns sampled frames and native app links. A brief event can fall between overview samples, and a similarity score is a ranking signal rather than a probability.