If you ask for something no tool covers, or that needs a permission you can’t or won’t give the connection, the assistant doesn’t just stop. It says Pixelbase can’t do that right now and offers to send the Pixelbase team a request.

You always see it first. The assistant drafts the message from your conversation, shows it to you, and sends it only once you say yes. It then tells you it went. submit_contact_inquiry never runs on its own.

Where it goes. To Pixelbase — not to your company and not to your customers. It lands in the same inbox as the Contact Us form on our website, and the team is notified. It arrives marked as coming through MCP, with who sent it, from which app and for which company (Pixelbase records those itself from your connection), and what the assistant said it was attempting.

What it takes. The inquiries:write permission, which every company’s baseline includes — take it out of the baseline under Settings → Company → Permissions to stop your people sending requests. Your name, email and phone default to your own profile. A message is at most 2,000 characters, and a connection can send 5 an hour (10 a day per person).

For clients. A failed tool answer that has somewhere to send you carries a nextSteps list — in the result’s structuredContent, and as a short list under the message for clients that show only text. Each step is { kind: 'submit_request', tool: 'submit_contact_inquiry' }, or { kind: 'approve_permissions', url, scopes } where a link that grants a missing permission exists; the link, when there is one, comes first.