Video MCP Prompt Examples for Creation, Clipping and Scheduling

Exact copy and a useful carousel

1. Render a message you have already written.

Use Prolifik and the brand I select from my account. Create one Text Story with this exact text: “[approved copy]”. Use download delivery and no music. Preserve the wording. Return the job ID, piece ID and finished media after rendering. Do not publish.

Supply the actual copy and confirm the brand. The expected tool intent is a content job using the Text Story format, with one customTexts entry containing text. Completion means a playable returned video containing the approved message. It does not mean the assistant has rewritten the text attractively in chat.

2. Render a carousel with a known slide count.

Create one carousel for [brand], using these five approved slide strings in order: [slides]. Treat the five slides as one content piece. Return the script and all rendered media that actually exist. Keep it as a draft for review.

Here, one customTexts entry contains a slides array. Inspect the returned slide order and whether you have images, a video or both. The carousel production guide helps adapt the copy to timed playback when needed.

Use a source without inventing footage

3. Base a new draft on an existing analyzed page.

List the sources for [brand] and retrieve the analysis for [selected source]. Draft one short message based only on its supported claims. Show me the text and the source passages it relies on. After I accept the text, create one B-roll video for download.

This separates source retrieval from creation. The assistant should confirm that the source analysis is ready, then use the agreed text in the content job. If the page cannot support a claim, remove the claim or obtain better evidence. Attractive background footage does not repair an unsupported statement.

4. Use a real app recording for Hook & Demo.

Find my uploaded recording [recognizable name] in the Prolifik media library and check that it is usable for Hook & Demo. Prepare one piece using that exact demo recording and the approved hook copy [text]. Use a library hook unless I supply my own. Return a download draft.

The required input is an actual eligible recording, represented by its media key. The talking-head content format requires demoVideoKey. Its source audio is replaced by library music or silence, so use a different workflow if original narrated audio must be preserved. A product URL alone is insufficient.

Extract a thought and correct a specific piece

5. Clip one complete explanation from a recording.

Check my clip-minute allowance, then request candidate clips from [owned finished recording URL]. Focus on the explanation of [specific subject]. Preserve the setup and any qualification that changes the meaning. Use vertical framing and readable captions. Keep the job ID and return the actual clips for review.

The assistant should use the clipping tools and retrieve the job's results. Selection guidance cannot guarantee perfect boundaries or transcription. Watch the first and final thought, then verify names in the captions. Clip boundary guidance is useful for the review.

6. Correct one rendered carousel without rewriting the batch.

Retrieve piece [piece ID]. Replace its slides with this corrected complete array: [slides]. Keep the intended logo setting. Queue the replacement render and return that piece once it is ready. Do not change the other pieces or any publication records.

This maps to a piece edit, not creation of a new batch. Give the full desired slide sequence so an omitted slide cannot be mistaken for an intentional deletion. Completion means the replacement render has finished and shows the correction.

Reconcile results and schedule an accepted video

7. Inspect a batch without generating more work.

Retrieve content job [job ID]. Match each returned piece to its script. List ready media and failures separately, including the actual failure reason when supplied. Do not retry, regenerate or publish anything. Tell me which requested messages lack a ready output.

This is a read-only review request. It should use retrieval tools, not spend more renders in an attempt to make the report look complete. The useful deliverable is a mapping from planned messages to actual files.

8. Schedule a particular accepted piece.

Retrieve the reviewed piece [piece ID] and list the connected destinations for its brand. Use the account I select. Show the approved caption and [date, time, named time zone] as an exact timestamp. After my confirmation, schedule that piece and return the publication identifier and stored time.

Publishing changes an external account, so preserve the client's confirmation step. The account must already be connected in Prolifik. Later, inspect the publication record to distinguish a scheduled item from one that actually posted.

Adapt the contract, not just the adjectives

When a prompt gives an unsatisfactory result, identify the missing decision. “More professional” may mean fewer words, a different background or an incorrect brand. Name the specific change and retrieve the resulting artifact again.

The MCP and API comparison explains the job state that remains behind these conversational requests. Keep the source, selected identifiers and completion condition in the request, then judge success by the returned media or publication record. That makes the prompts reusable without pretending every tool or client can perform the same action.