MCP image generation for social media managers

Treat MCP as a controlled concept-production loop

For a social media manager, the useful MCP pattern is not “make content” in one opaque request. Define one campaign idea, choose a working canvas, inspect the current image catalog, confirm the credit balance, and ask for one concept with an explicit review target. The published OfflineCreator package exposes separate tools for model discovery, balance lookup, generation, status polling, and completed-output retrieval, so a client can keep those checkpoints visible instead of hiding them inside a single prompt.

A practical brief names the audience, message, focal subject, palette, composition, exclusions, and copy-safe area. Prepare three composition briefs for `1:1`, `9:16`, and `16:9`, but do not assume that changing the ratio alone creates a finished placement. The current `generate` schema accepts those three aspect-ratio values and defaults to `16:9`; it does not promise that one composition will survive every crop or that generated lettering will be usable.

Discover

list_models and get_creditsRead the current model IDs, stated fits, per-run credits, and account balance before requesting media.

Generate

One brief, one ratio, one approvalKeep each paid attempt tied to a specific placement hypothesis and an observable pass condition.

Retrieve

wait_generation, get_generation, download_outputTrack the generation ID and retrieve the completed output rather than starting an unrecorded replacement.

Put the cost decision before every iteration

OfflineCreator's current image catalog lists FLUX Schnell at 1 credit for fast iteration, Recraft V3 at 5 credits for brand graphics, and FLUX 1.1 Pro Ultra at 8 credits for photoreal detail. Those labels are catalog positioning, not an independent quality ranking. Start with the model whose stated fit matches the question you need to answer, and record the displayed cost as a dated input rather than a permanent price.

Turn iteration into a bounded comparison. Approve a maximum attempt count and credit ceiling, hold the ratio and core brief steady, then change one variable per attempt: subject placement, background density, palette, or negative-space instruction. The Studio product page says models, provider, aspect ratios, and credit costs remain visible before generation, and that MCP uses the same account balance and published model costs as the web experience. Read those values at execution time; this research did not access an account or observe a live balance.

Fast composition test

FLUX Schnell · 1 creditUse the catalog's fast-iteration route to test hierarchy or crop strategy, not to claim final quality.

Graphic direction

Recraft V3 · 5 creditsUse when the catalog's brand-graphics positioning matches the brief.

Detailed still

FLUX 1.1 Pro Ultra · 8 creditsUse when the catalog's photoreal-detail positioning is relevant and the higher attempt cost is approved.

Plan square, vertical, and landscape as separate compositions

Use `1:1` for the square proof, `9:16` for the vertical proof, and `16:9` for the landscape proof because those are the three values accepted by the version-pinned MCP generation schema. That is an interface fact, not a universal publishing specification. A social platform can impose different dimensions, crop behavior, overlays, file requirements, or safe zones, so validate the chosen destination before export.

The ratio matrix should preserve the campaign idea while allowing the layout to change. In the square version, protect the center and keep the focal subject readable at a small preview. In the vertical version, leave room where platform interface and captions may compete with the artwork. TikTok's June 2026 Non-Spark in-feed video-ad specification recommends `9:16` vertical, also accepts `16:9` horizontal and `1:1` square, and says safe zones vary with dimensions, caption length, and added formats. That narrow video-ad guidance is not a universal organic-image rule. In the landscape version, use width intentionally instead of stretching the square scene. Add exact campaign copy later as editable type; no source reviewed here establishes reliable spelling or typography inside generated pixels.

1:1 square proof

Centered hierarchyReview the focal subject, logo-free art, and reserved copy area at feed-preview size.

9:16 vertical proof

Tall composition with protected edgesCheck the actual destination's current overlays and safe-zone guidance before publishing.

16:9 landscape proof

Wide composition with deliberate side spaceUse the wider frame for a distinct layout, not as an automatic crop of the square output.

Review a proof set before requesting more variants

Create one proof sheet with the exact prompt, model ID, displayed credit cost, ratio, generation ID, unedited output, and decision for every attempt. Review message accuracy, subject integrity, unintended text or marks, composition at the target preview size, copy-safe space, and obvious crop risks. Mark each output accepted, rejected, or revise, and name the first failing criterion. This is an editorial QA record, not a benchmark or a performance test.

If an attempt fails, revise the smallest responsible input. Move the subject when the crop is wrong, simplify the background when hierarchy is weak, or rewrite the exclusion when an unwanted element appears. Do not infer that a more expensive model will fix a composition problem. OfflineCreator states that failed generation jobs refund reserved credits automatically, but an output that completes and merely fails editorial review is not documented as a failed job; keep editorial iteration inside the approved budget.

Accept

Meets the written visual and crop checksRetain the source record and move to deterministic finishing with editable copy.

Revise

One named input can address the defectChange one variable and preserve the prior output for comparison.

Reject

Misleading, unusable, or outside the approved directionDo not publish or use the asset as evidence of campaign performance.

Choose the next guide from the production bottleneck

Return to the creator use-case directory when the output type is still undecided. Use the content-creator page when one approved source must become a coordinated package rather than three social-image concepts. Move to the CLI social-variant guide when the team needs a repeatable command-driven batch with its own manifest and review record. These links divide the next action by workflow, not by repeating the same search phrase.

Evidence and ownership boundary

This page owns the narrow question of using MCP to develop reviewed social-image concepts across square, vertical, and landscape working canvases, with current image-model choices and iteration costs. It does not claim automated publishing, platform API integration, campaign performance, deterministic reproduction, universal placement specifications, or finished typography. Consolidate it into the creator use-case directory if current ratios, image models, and iteration costs can no longer be supported.

Recent community coverage remains insufficient for product claims. The 2026-08-09 last30days run returned 69 items but was degraded: optional X credentials were not configured; Reddit ended partial after RSS feed failures; and the few social-manager MCP mentions were low-engagement practitioner or promotional anecdotes about third-party tools. They do not substantiate OfflineCreator ratios, catalog model costs, or social-team outcomes, and they are not evidence that practitioners are silent. Current claims rely on later first-party product, package, protocol, and narrowly scoped platform documentation.