Arcads MCP Setup Checklist: Approval, Batch, and Credit Guardrails

Arcads now documents a hosted MCP server for creating AI-generated ads from compatible MCP clients. A connection can be established quickly; the operational risk begins after connection, when a broad prompt can turn into an undefined batch, unexpected credit use, or assets that bypass review.

This checklist defines the smallest safe production contract for a first Arcads MCP workflow. Confirm the current prerequisite and connection steps in the official Arcads MCP guide before implementing it.

1. Write the finish line before connecting

Describe one review-ready outcome, not an open-ended creative goal. A workable finish line names the source material, output format, maximum number of requested assets, review owner, and the point at which the workflow must stop.

Example: From one owner-approved product brief and reference ad, request no more than five hook variations. Stop before publishing. Route every result to the creative operations owner for review.

2. Confirm six preflight decisions

1. Source rights: Who owns or has permission to use the product assets, reference creative, actor likenesses, scripts, and claims?
2. Output contract: What exact asset type, format, locale, and variation dimension may be requested?
3. Batch cap: What is the maximum number of outputs in one production run?
4. Credit ceiling: What maximum Arcads credit exposure may the operator authorize for the run?
5. Approval gate: Who approves the final inputs and cap before a credit-consuming action?
6. Exception owner: Who decides what happens when authorization, credits, source data, or output quality fails?

3. Connect without sharing credentials

Arcads currently documents the hosted server at https://mcp.arcads.ai and lists an Arcads account plus an active subscription with remaining credits as prerequisites. The account owner should add the custom connector and complete authorization in the supported client. Passwords, API secrets, and tokens do not belong in planning documents, intake forms, chat messages, or run logs.

4. Use a bounded operating contract

PURPOSE
Create one review-ready asset set. Do not publish or spend.

ALLOWED INPUT
One owner-approved product brief and reference asset.

BATCH BOUNDARY
Maximum 5 requested outputs. Apply the client-defined Arcads credit ceiling.

APPROVAL
Named operator approves the final input, claims, and batch cap before generation.

STOP CONDITIONS
Stop on missing approval, insufficient credits, authorization failure, out-of-scope input, or cap reached.

HANDOFF
Stage every result for human review and record useful output identifiers without credentials.

5. Test the control points, not only the happy path

• Missing approval creates no production generation.
• The requested batch cannot exceed the written cap.
• Prohibited or private inputs stop before generation.
• Authorization and credit failures return an actionable stop state.
• Every generated asset reaches the named reviewer.
• Publishing and media spend remain disabled.

6. Keep the first pilot narrow

A useful first pilot covers one compatible MCP client, one Arcads account, one approved playbook, connection validation, one capped test, acceptance checks, operating notes, and an operator walkthrough. It should exclude Arcads subscription costs, generation credits, claims or legal review, usage rights, automatic publishing, media spend, multiple brands, and unbounded batches unless those items receive their own explicit scope.

Build the playbook before the run

Use the free Arcads MCP Playbook Builder to generate a copyable version of the connection checklist, approval gate, batch and credit boundaries, stop conditions, and acceptance tests in your browser. No account or credentials are required.

For implementation, FlowPatch offers a fixed $299 Arcads MCP setup pilot for one client and one approved playbook. The mini-audit and written-scope request are free and create no charge.