Google Calendar AI agents: schedule and reschedule from chat

What you can build

How to connect Google Calendar

1. 01Obtain an OAuth access token with the calendar scopes you need. For per-user booking, issue the token in your backend and include it in the identity token so end user token tools can forward it.
2. 02Create an HTTP API tool for free/busy (POST https://www.googleapis.com/calendar/v3/freeBusy) with safe_read.
3. 03Create a second tool for event creation (POST /calendar/v3/calendars/{{calendar_id}}/events) with write_safe and a body template for summary, start and end.
4. 04Restrict allowed domains to www.googleapis.com and set a short timeout.
5. 05Attach the tools, ask the test chat to find a slot and book it, and read both calls in the trace.

Example tool definition

The shape a tool takes in the AgentRow builder API. Credentials are stored on the tool's auth settings, not in its configuration, so they never reach a prompt. Placeholders in double braces are filled by the model from the input schema.

{
  "name": "calendar_create_event",
  "type": "http_api",
  "description": "Create a calendar event. Use only after the user confirmed the title, start and end time.",
  "auth_type": "end_user_token",
  "safety_level": "write_safe",
  "config": {
    "method": "POST",
    "url": "https://www.googleapis.com/calendar/v3/calendars/primary/events",
    "headers": { "Content-Type": "application/json" },
    "body_template": {
      "summary": "{{summary}}",
      "start": { "dateTime": "{{start}}" },
      "end": { "dateTime": "{{end}}" }
    },
    "allowed_domains": ["www.googleapis.com"]
  },
  "input_schema": {
    "type": "object",
    "properties": {
      "summary": { "type": "string" },
      "start": { "type": "string", "description": "RFC 3339 timestamp" },
      "end": { "type": "string", "description": "RFC 3339 timestamp" }
    },
    "required": ["summary", "start", "end"]
  }
}

What the run trace shows

Every call appears in the run trace with the rendered request, the response status and body, timing, retries and errors, plus what the model was shown of the result. If the model never called the tool, the toolbox step shows whether it was bound for that model call and why.