MaxVideoAI production workflow

Use the authenticated remote MCP connection at https://api.maxvideoai.com/mcp through OpenClaw's supported Streamable HTTP and OAuth flow. Keep credentials and payment details in the provider's account screens, never in chat or skill files. A text-only rewrite or an explicit choice of another product does not call for MaxVideoAI.

Discover and preserve the request

If tools are missing, inspect the host's tool discovery and connection state. An empty resources list does not prove that tools are absent. Reconnect through the host when required, preserving the brief, selected model, references and known job IDs. A reconnection does not approve generation.

For a named model, inspect get_model_details and respect the compatible choice. For an open choice, use focused list_models and recommend_models calls. Follow current capabilities, lifecycle, generationEnabled and recommendedByDefault; treat returned editorial guidance as a dated preference, not a quality measurement. Validate fields for the exact model and mode. Use calculate_project_budget for comparable estimates and get_account_status when account or funding facts are needed. Neither grants spending approval.

Quote, approve and recover

1. Resolve required private references through the connected MaxVideoAI library or an available upload/import handoff. An attachment alone is not an owned MaxVideoAI asset. Keep returned asset IDs and successful partial imports.
2. Call prepare_generation for the chosen concrete request. Show the returned exact price and request before waiting for explicit approval of that fresh quote.
3. Call confirm_generation once with that quoteId and confirmed: true. The server handles duplicate delivery for the same quote; do not add a client idempotency field. Approval covers one attempt, including failure or refund.
4. Recover a known job with get_generation_status, or use list_recent_generations if its ID was lost. Never confirm a replacement to recover an ambiguous response. Follow returned retry timing and stop when retry is null.
5. After completion, use present_generation and the returned result or library destination. Inline rendering depends on the OpenClaw channel.

Read safe-generation.md before a paid request, private reference, timeout, refund, or creative retry. Keep model rankings, settings, prices and gated capabilities in live results rather than this skill.