n8n calendar booking that doesn't double-book.

Add one HTTP Request node and point it at AgentDraft. No community node to install — it is a plain authenticated HTTP call, so it works on n8n Cloud and self-hosted alike.

• Method & URL — POST https://api.agentdraft.io/v1/bookings.
• Authentication — Generic Credential → Header Auth. Name Authorization, value Bearer avs_live_… (an agent key from the dashboard). n8n stores it encrypted, off the canvas.
• Header — add Idempotency-Key so a re-run of the same workflow never books twice.
• Body — JSON, shown below.

{
  "start": "2026-05-12T16:00:00Z",
  "end":   "2026-05-12T16:30:00Z",
  "mode":  "commit",
  "metadata": { "title": "Discovery call · Acme" }
}

A 201 returns the booking_id, status, and an audit_event_id. A 409 means another agent won the slot — its body carries winning_agent_id and winning_agent_priority so the next node can pick another time. Set the node's Never Error response option so the 409 flows into an IF branch rather than failing the execution.

Import the ready-made workflow and wire your own trigger in front of it. It ships the booking call plus the conflict branch already split:

Manual Trigger
   └─ Booking details (Set)
        └─ Commit booking (HTTP Request → /v1/bookings)
             └─ Committed?  (IF statusCode == 201)
                  ├─ true  → Booked
                  └─ false → Outranked — try another slot

In n8n: Workflows → ⋯ → Import from File, then open the Commit booking node and select your AgentDraft Header Auth credential. Download agentdraft-n8n-workflow.json.

n8n makes it trivial to wire a Google Calendar node into a workflow — and that is exactly where the trouble starts at scale. The Google Calendar node writes the event directly, so two workflows (or two runs of the same workflow, or an n8n AI agent racing a human's assistant) that both see a free slot both succeed in writing. The human gets overlapping events. The multi-agent calendar collision is the predictable consequence.

Routing the booking through AgentDraft first moves the race off the calendar and onto the coordination layer. The conflict engine resolves it at the storage layer — a single TransactWriteItems with conditional writes per time bucket — and only the winner's write reaches the underlying calendar. Keep your Google Calendar node for reads and human-facing flows; let AgentDraft own the contended write.

Inside an AI Agent node, attach the same call as an HTTP Request tool. The agent decides when to book; AgentDraft decides which agent wins. Because a lost race comes back as a structured 409 rather than an exception, the model reads the winner's priority and proposes the next slot — the conflict becomes a normal turn in the conversation, not a failed run.

• Why AI scheduling agents collide — and what a coordination layer looks like — the protocol-level case for AgentDraft.
• Calendar API for AI agents — the full HTTP surface the HTTP Request node calls.
• The MCP server — the same coordination from any MCP client.