Approval workflows for MCP campaign assets

Separate generation approval from campaign approval

An MCP campaign-asset approval workflow should make two decisions visible. The first authorizes a bounded generation attempt: one model, one prompt, one ratio, one displayed cost, and one named reviewer. The second accepts or rejects the resulting file for a specific campaign placement. Passing the first gate never makes the output ready for delivery. It only creates a candidate that still needs claims, rights, brand, accessibility, disclosure, and destination checks.

The published @offlinecreator/mcp 0.1.2 package supports a generation lifecycle through tools for model discovery, credit lookup, generation, image upload, status checks, waiting, output retrieval, cancellation, and recent-job listing. It does not document campaign comments, legal sign-off, role-based approval routing, asset version approval, or publishing tools. Keep those decisions in the governed review system the team already uses instead of presenting this page as a built-in collaboration feature.

Gate 1

Authorize one generationFreeze the prompt, model, ratio, source input, displayed cost, attempt ceiling, and operator before spending credits.

Gate 2

Approve one campaign useReview the retrieved file against a named placement and record accept, revise, or reject.

System boundary

Use an external review recordDo not infer shared workspaces, reviewer roles, comments, or publishing controls from the MCP generation tool set.

Place human checks before spend and before delivery

Before generation, confirm that the source image is authorized for this use, the brief does not request an unsupported factual depiction, the live catalog choice fits the intended workflow, the account has enough credits, and the operator accepts the displayed cost. OfflineCreator says its MCP interface shows the selected model, provider, aspect ratio, and credit cost before a generation starts, and uses the same account balance and published model costs as Studio. Re-read those values at execution time rather than copying them into a permanent template.

Before delivery, inspect the actual output rather than approving the prompt. Verify the intended subject and product, compare every objective claim with its evidence, check logos and visual identity against current brand files, identify accidental text or marks, confirm crop and contrast in the real placement, and decide what disclosure accompanies the asset. The FTC's advertising-substantiation policy requires a reasonable basis for express and implied objective advertising claims before dissemination, so a visually plausible output cannot serve as its own evidence.

Pre-spend

Rights, model, provider, cost, and stop ruleApprove the exact request and a maximum attempt count before a paid tool call.

Pre-delivery

Claims, brand, placement, disclosure, and provenanceApprove the retrieved file only after checking what viewers will actually see and infer.

Escalate

Regulated, sensitive, or ambiguous claimsSend the asset to the responsible legal or compliance owner; this checklist is not legal advice.

Use a seven-part campaign asset review card

Attach one review card to each candidate, not just to the campaign folder. Record the generation ID, unedited output, prompt, source inputs, model and provider, displayed cost, intended placement, reviewer, and decision. Then answer seven questions: Is every objective claim supported? Are source-media rights documented? Does the output match the approved product and brand system? Is required copy editable and readable? Does the placement need a sponsorship or other disclosure? Is AI provenance represented accurately? Is there a named person authorized to release the file?

For endorsements, the FTC advises that a material connection to a brand should be obvious and that the disclosure should appear with the endorsement where people will notice it. In video, the agency says the disclosure should be in the video rather than only in its description. That guidance does not mean every AI-generated campaign asset needs the same label; the reviewer must identify the actual relationship, medium, audience, and applicable rules.

Claims and depiction

Supported and not misleadingCheck express copy, implied messages, product appearance, demonstrations, comparisons, prices, and performance cues.

Rights and brand

Documented sources and current identityRetain licenses or permissions and compare logos, colors, packaging, people, and locations with approved references.

Delivery and disclosure

Placement-ready with a release ownerTest crop, readable copy, accessibility, sponsorship language, AI provenance, destination rules, and final authorization.

Reject the polished asset when its evidence trail is weak

Consider a product image that looks on-brand but invents a feature label on the package. The prompt may have been approved and the composition may be attractive, yet the candidate fails because viewers could interpret the invented label as an objective product claim. Mark the file rejected, preserve it with the reason, and return to an approved product photograph, a corrected compositing workflow, or a new generation whose acceptance criteria prohibit packaging text. Do not quietly paint over the label and lose the decision history.

Apply the same discipline to ownership language. The U.S. Copyright Office reports that copyright does not extend to purely AI-generated material and that prompts alone, with current generally available technology, do not provide sufficient human control; human-authored expression, creative arrangement, or modification may still be protected case by case. A reviewer should record human contributions and source permissions without promising that a prompt or generation receipt proves exclusive rights.

If the delivery carries Content Credentials, describe them narrowly. C2PA 2.2 provides a technical structure for signed, tamper-evident provenance assertions, but its own principles say the specification should not judge whether provenance data is good or bad. A valid manifest can help inspect origin and edits; it does not prove that an advertising claim, license, or brand approval is true.

Continue from the next unresolved campaign decision

Use the creator-use-case directory when the asset type is still undecided. Move to short-form video concepting when the approval problem is a bounded motion brief, or to client-deliverable packaging when accepted concepts need files, caveats, costs, and review status organized for handoff. These paths keep this page focused on the approval decision rather than duplicating generation or delivery instructions.

Evidence boundary for campaign approval claims

This editorial workflow does not imply that OfflineCreator supplies campaign-management or team-approval features. The public versioned MCP tool list supports the generation boundary described here, while the current pricing page does not document a Team or Enterprise tier or reviewer seats. Keep collaboration, legal review, release authority, and publishing in systems that explicitly govern those decisions.

Recent community evidence is not used to support product behavior or practitioner adoption on this page. One one-point Reddit post described an experimental proposal, human approval, authorization receipt, linked execution, and event sequence. That is a single uncorroborated anecdote, not evidence of a common campaign-asset practice or OfflineCreator capability.