MCP shot concepts for video editors

Use MCP to create shot material, not to replace the edit

For video editors, the narrow use of MCP is to create one short insert or motion reference after the edit has identified a specific gap. Define the shot's purpose, subject, framing, movement, aspect ratio, continuity anchors, and rejection criteria before requesting a generation. Retrieve the completed file, then bring it into the editor as candidate media. The timeline still owns selection, trimming, pacing, transitions, color, sound, captions, graphics, rights review, and final export.

OfflineCreator's current launch catalog identifies Kling 2.6 Pro as text-to-video, Kling 2.6 Pro Motion as image-to-video, and Veo 3.1 Fast as text-to-video. These workflow labels help route a bounded shot request, but they do not establish that a generated clip will match adjacent footage, preserve a person's identity, hold an exact product, or arrive ready for a locked sequence. Treat every result as a proof that must survive an editor's continuity and technical review.

Plan around short clip units without misstating the limits

The useful planning numbers are five seconds for a default Kling request and up to eight seconds for the documented Veo 3.1 Fast choices, but neither number should be published as a universal model limit. fal currently documents Kling 2.6 Pro text-to-video and image-to-video with a five-second default and five- or ten-second options. Google documents Veo 3.1 Fast with four-, six-, or eight-second lengths. Provider capabilities can be broader than the controls exposed by a particular integration.

Write each request as one visual beat that can enter and leave cleanly: one reveal, one camera move, one environmental transition, or one restrained animation of an approved still. Specify usable handles in the concept, but verify the actual returned frames instead of promising exact pre-roll or post-roll. If the edit needs dialogue continuity, multiple angles, precise action timing, or a long unbroken performance, treat generation as reference or source fragments and solve the sequence in the editing system.

Kling 2.6 Pro

Five or ten secondsfal's current text-to-video and image-to-video schemas default to five seconds and also allow ten.

Veo 3.1 Fast

Four, six, or eight secondsGoogle's current Fast documentation lists these three output lengths.

Editorial implication

Design one beat per generationDo not assume that a prompt can produce a finished multi-shot sequence.

Keep generation tools separate from editing tools

The published @offlinecreator/mcp 0.1.2 README lists tools for model discovery, credit balance, starting a generation, uploading an image for image-to-video, checking and waiting on job state, retrieving a completed output, cancelling a reserved job, and listing recent generations. It does not document timeline assembly, source trimming, ripple edits, transitions, multicam synchronization, color correction, audio mixing, caption authoring, title design, proxy management, or final delivery as OfflineCreator MCP tools.

The version-pinned server implementation reinforces that boundary. Its generate input contains a model ID, prompt, optional aspect ratio, and optional wait flag; it exposes no duration field and no edit-decision list or timeline arguments. Do not invent a duration control because the underlying provider supports one, and do not describe a returned MP4 as an edited deliverable. Download or stage the completed output, preserve its generation record, and perform deterministic finishing in the NLE or another purpose-built media pipeline.

Write a shot card that survives the handoff

A useful shot card names the editorial job first: for example, a vertical five-to-eight-second atmosphere insert between two interview answers, with no recognizable person, no readable signage, a slow forward move, cool dawn light, and enough visual calm for a live caption. Add the intended sequence, target frame ratio, continuity references, prohibited elements, source-rights status, model choice, expected cost, reviewer, and a stop rule. This turns an open-ended generation request into a testable edit decision.

After generation, append the generation identifier, returned duration and dimensions, file checksum, retrieval date, provider disclosure, and review disposition. Record observable defects such as subject drift, discontinuous motion, unwanted text, unstable geometry, abrupt first or last frames, audio that will be discarded, or insufficient handles. The card should make rejection useful: an editor can decide whether to revise the prompt, select a different workflow, use the result only as motion reference, or abandon generation and source licensed footage.

Choose text-to-video, image-to-video, or a conventional source

Choose the workflow from the acceptance criterion, not from novelty. Text-to-video is the safer concept route when the shot can be fictional and self-contained. Image-to-video is relevant when the team has an authorized still and wants to test restrained motion, but the upload path is not a preservation guarantee. Use licensed footage, original capture, compositing, or deterministic animation when the sequence depends on documentary truth, exact branding, repeatable typography, a recognizable performance, or frame-accurate continuity.

One recent draft GitHub pull request offers an anecdotal architecture for this separation: it proposes an editor-neutral clip plan and explicit media materialization while stating that the research server is not becoming an editor. That is a useful design example, not evidence of broad practitioner adoption or a completed standard. Its draft status and single-repository scope mean this page does not generalize it into a claim about how video teams currently work.

Text-to-video

The composition can be inventedUse for a self-contained atmospheric insert where exact identity and continuity are not acceptance criteria.

Image-to-video

An approved still should anchor the conceptUse the documented upload path, then inspect every frame for drift from the source.

Licensed or captured footage

Authenticity must be exactPrefer a governed source when the shot must prove a person, place, product, event, claim, or action.

Continue from the editor's next unresolved decision

Use the creator-use-case directory when the deliverable is still undefined, the reference-led workflow when an authorized still is central to the motion test, and the performance-creative page when the team needs controlled concept variants rather than a narrative insert. These routes keep this page focused on a specific handoff: generate a short candidate shot, preserve its evidence, and finish the sequence in an editor.

Evidence boundary for MCP video-editor claims

This draft does not claim that OfflineCreator edits timelines or that five and eight seconds are the only durations supported by the underlying models. Current provider documentation instead shows five or ten seconds for Kling 2.6 Pro and four, six, or eight seconds for Veo 3.1 Fast, while the public OfflineCreator MCP 0.1.2 generate schema exposes no duration input. The page therefore treats those lengths as shot-planning evidence and tells readers to inspect the live integration rather than promising an unavailable control.

Recent community coverage was degraded and mostly unrelated. The single draft GitHub handoff cited here is explicitly presented as an anecdotal design example; it does not establish adoption, compatibility, reliability, speed, quality, or customer outcomes. No OAuth connection, paid generation, image upload, returned clip, NLE import, duration observation, cancellation, refund, or output-retrieval test was performed. Keep this workflow as a researched proposal until a reviewer verifies it against a current account and a real editing environment.