What the assistant may do — and what it may not — Agent control (MCP) — VidVertex Docs
Docs/Agent control (MCP)/What the assistant may do — and what it may not

What the assistant may do — and what it may not

The same guardrails as every unattended context apply (see Automation):

  • Posting is dry-run by default. schedule and run_assembly show exactly what would post; a real upload happens only after you explicitly approve it in the conversation — or per post in the app's review queue, where every proposed post from an agentic dry run waits with its preview and texts.
  • Posting is asynchronous and duplicate-safe. schedule returns a run id and keeps posting in the background; upload_status reports the state — even after a restart, where an interrupted run says exactly which posts were made. Exact duplicates are skipped and reported; posting the same file again on purpose has to be asked for.
  • Deleting requires your confirmation. Every delete call is refused until the assistant passes a confirmation flag it may only set after you said yes. Approving review-queue entries through a tool is gated the same way.
WARNING

The Google Drive sign-in has no tool on purpose, and that has a visible consequence: an assistant's Metricool run ends with "Google Drive is not connected" until you connect Drive once in Settings → Upload settings.

  • Rate limits hold. Approved posts respect the same per-account limits as every other part of the app.
  • API keys stay out. The assistant can see whether a key is configured, never the key itself — and it cannot enter one. It also cannot switch AI providers or change the guardrails themselves.
  • Some things have no tool on purpose: changing the data folder, installing an update, license activation, the Drive sign-in above, and the extension export itself.