Plan multi-format content packages with MCP

Plan one package, not one master file

A multi-format content package starts with one approved message and separates it into deliverables before generation. Write a package matrix with a row for every placement: purpose, medium, aspect ratio, target dimensions, duration where relevant, subject position, safe area, copy treatment, motion requirement, source rights, reviewer, and final assembler. The shared brief should hold the campaign idea, factual claims, visual anchors, prohibited elements, and approval criteria. Geometry-specific instructions belong in each row.

Do not assume that a wide generation can be cropped into every other placement. A 16:9 composition may put the subject or product where a 9:16 crop removes it, while a square still may need a different camera distance to preserve both the focal point and room for live copy. Generate or compose each materially different geometry intentionally, then review the package for family resemblance. Consistency means shared message and art direction, not identical pixels.

Shared

Message, claims, visual anchors, rights, and review rubricFreeze these fields before splitting the brief into format-specific rows.

Per deliverable

Placement, medium, ratio, dimensions, safe area, and assembly ownerA row is not ready for generation while any required delivery field remains unknown.

Use destination specifications as the ratio source

Treat a platform's current delivery specification as the source of truth, not a remembered list of social ratios. Google Ads' current Performance Max image guidance requires horizontal 1.91:1 and square 1:1 assets, lists 1200 × 628 and 1200 × 1200 as recommended sizes, and supports an optional vertical 4:5 asset at 960 × 1200. A package intended for that campaign therefore needs at least the two required geometries; a generic wide, square, and vertical label is not precise enough.

The same Google Ads guidance also states that content needs to be in the center 80% of the image as the safe area that will not be cut off across device screen sizes. Record that verified value in the Performance Max still rows. For other destinations, record safe zones only after retrieving each destination's current first-party specification directly. If that source cannot be retrieved, leave the values unresolved rather than copying a remembered percentage or a search-result snippet.

Performance Max stills

1.91:1 and 1:1 required; 4:5 optionalUse the cited current sizes in the delivery row and recheck the platform specification before export.

Performance Max safe area

Center 80% of the imageCited from current Google Ads Performance Max image guidance; re-verify before export.

Match every row to a model-supported geometry

Build a support check before any paid generation call. For each row, record the exact provider, model identifier, interface, accepted aspect ratio, output resolution, duration options for motion, required inputs, and the documentation retrieval date. Mark a row blocked when its required geometry is absent. Do not silently substitute the nearest ratio or infer support from another model with a similar name.

Google's current Vertex AI documentation gives a concrete model-specific boundary: Veo text-to-video accepts 16:9 and 9:16. That supports a wide and vertical motion pair for that documented interface, but it does not support a claim that the same call produces a square motion asset. If the package also needs 1:1, plan a separately supported generation, a reviewed crop or padded composition in downstream software, or an explicit no-deliverable decision. This example does not establish that Veo is available through OfflineCreator Studio.

Supported

Exact model, interface, ratio, and retrieval date recordedProceed only when the required geometry appears in current first-party documentation or tool metadata.

Unsupported

Choose another model, downstream composition, or no deliverableNever relabel a near-ratio output as if it met the destination specification.

Use MCP for bounded calls and an auditable handoff

MCP can carry a package plan as explicit tool arguments and return tool results, but the protocol does not turn a set of generation calls into a finished campaign package. The current MCP tools specification defines discoverable tools with input schemas, optional output schemas, and structured or unstructured results. It also recommends a human in the loop with the ability to deny tool invocations. Use those boundaries to make each generation row inspectable and approval-gated.

For each call, preserve the row identifier, model identifier, prompt version, declared ratio, source input reference, returned generation identifier, status, selected output, rejection reason, and reviewer. Validate structured results when a tool provides an output schema, and treat returned media or links as inputs to the package record. A successful tool result means the call returned; it does not mean the crop, claim accuracy, brand fit, safe zones, or cross-format continuity passed review.

Before invocation

Show row, model, ratio, input, and expected costRequire confirmation when the call spends credits or uses an uploaded source.

After invocation

Attach result, status, selection, and review decisionKeep rejected attempts in the record so the package is not presented as a frictionless one-call outcome.

Make downstream assembly an explicit boundary

OfflineCreator Studio's published MCP 0.1.1 package lists tools for model discovery, credit balance, generation, source-image upload, generation status and waiting, output download, cancellation, and recent-job listing. It does not list timeline editing, multi-asset compositing, responsive cropping, typography, caption layout, audio mixing, or delivery encoding tools. Plan those operations in a separate editor, renderer, or media pipeline instead of implying that an MCP generation result is an assembled content package.

Create an assembly manifest for each selected output. Include the source generation identifier and hash, destination dimensions, crop or pad rule, copy and font source, logo placement, safe zones, caption or audio files, duration, frame rate, codec, output filename, and responsible tool. FFmpeg's official filter documentation includes filtergraphs and operations such as crop, scale, overlay, and concat, which can implement part of a video assembly plan. Their availability does not choose a good crop or approve a composition; keep a rendered proof and human decision beside every export.

Generation boundary

Downloaded still or clip plus its job recordThis is source media for assembly, not evidence that delivery specifications have been met.

Assembly boundary

Crop, layout, copy, captions, audio, timing, and encodingAssign these operations to a named downstream tool and reviewer.

Choose the next page by the unresolved decision

Use the creator-use-case hub when the team has not selected a media outcome. Move to creative proof-set operations when the open question is how many controlled attempts to authorize and how to record a stop-or-scale decision. Use short-form video concepting when the package already calls for a vertical motion row and the next task is defining its hook, camera motion, duration, and editor handoff. These links narrow the next decision rather than repeating this page's ratio matrix.

Evidence limits and package boundary

Recent community evidence was insufficient for publication claims. The last30days v3.18.4 run returned 78 dated items across Digg, GitHub, Hacker News, Instagram, TikTok, YouTube, grounding, and partially covered Reddit, but classification found zero relevant items for multi-format content packaging with supported ratios and an assembly boundary. Visible clusters were MCP protocol news, promotional product posts, general MCP usage threads, and unrelated repositories. X was skipped-unconfigured; Reddit ended partial; and arXiv, Polymarket, and Techmeme returned no results. Engine stderr recorded a RedditRSS feed future failure without an HTTP status code. Those access and relevance gaps do not establish that the affected sources had no relevant discussion, so this page makes no claim about adoption, preferred tools, time savings, quality, reliability, or business outcomes.

The factual scope is narrower: current Google Ads guidance supports the cited still-delivery geometries and the center-80% safe area; current Vertex AI documentation supports the cited Veo ratio pair; the MCP specification supports schema-described tool calls, results, and a human-control recommendation; the published OfflineCreator MCP 0.1.1 package supports the listed generation workflow boundary; and FFmpeg documents downstream media filters. No generation, package assembly, platform upload, cost observation, or cross-format user study was performed. The workflow is an authored operating plan, not a promise that one prompt, model, or product creates a finished multi-format campaign.