skdul MCP server: book meetings directly from ChatGPT or Claude

You already use ChatGPT or Claude for a dozen tasks every day. Email drafting, research, code review, data analysis. But the moment someone says "let's find a time to meet," you leave the chat, open your calendar, squint at colored blocks, and copy-paste time options into a message.

That workflow is broken. Here's how to fix it in two minutes.

What skdul's MCP server gives your AI assistant

skdul's MCP server exposes 16+ scheduling tools that your AI assistant can call directly:

• Browse booking pages — discover someone's available events by username.
• Check availability — see open time slots for any date range, in any timezone.
• Score and rank slots — the find_and_book_best_slot tool evaluates 100+ slots and returns the top options with a 0–100 quality score.
• Create bookings — confirm a meeting with one command. Both parties get email notifications.
• Manage your calendar — list upcoming meetings, reschedule, cancel, detect conflicts.
• Set preferences — tell the system you prefer mornings, avoid Fridays, or need 15-minute buffers.

These aren't stripped-down endpoints. Your AI assistant gets the same capabilities as the web dashboard. The difference is that it accesses them through conversation instead of clicks.

Setup: Claude Desktop

Open your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add this:

{
  "mcpServers": {
    "skdul": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/skdul"]
    }
  }
}

Restart Claude Desktop. The first time Claude calls a skdul tool, you'll authenticate via browser — sign in, authorize, done. That's the entire setup.

Setup: ChatGPT

ChatGPT's MCP integration works through its tool configuration. Add skdul as an MCP server in your ChatGPT settings under Tools → MCP Servers. The connection string is the same npx command. ChatGPT handles the rest — tool discovery, schema parsing, and authentication.

Real examples

Here's what scheduling looks like when your AI assistant has calendar access:

Book a meeting with someone

"Book a 30-minute call with Sarah next week. She's on skdul.ai/sarah. Mornings preferred."

The assistant browses Sarah's booking page, fetches her available slots for next week, scores them against your morning preference, and presents the best option. You say "book it," and the meeting is confirmed. Sarah gets an email. You get a calendar event. Total time: 15 seconds.

Check your day

"What do I have tomorrow? Any conflicts?"

The assistant pulls your bookings for tomorrow and runs conflict detection. You get a clean summary: three meetings, no overlaps, 2-hour focus block in the afternoon.

Reschedule intelligently

"Move my 2 PM meeting with Alex to Thursday. Same time if possible, otherwise morning."

The assistant finds the specific booking, checks Thursday availability, and reschedules. Alex gets a notification with the updated time. No "does this work for you?" emails.

Batch-book a week of 1:1s

"I need to schedule 30-minute 1:1s with each of my five direct reports next week. Afternoon slots, no back-to-back."

The assistant finds non-overlapping afternoon slots across all five calendars and presents a complete schedule. One approval books all five meetings.

Protect your focus time

"Set a preference: no meetings before 10 AM. I need mornings for deep work."

The assistant calls set_scheduling_preference with a time preference rule. From now on, all agent bookings and slot scoring respect your morning focus block.

How slot scoring works

When the assistant calls find_and_book_best_slot, skdul doesn't just return a random open time. It scores every available slot on five dimensions:

1. Time-of-day preference — if you prefer mornings, a 9 AM slot outscores a 4 PM slot.
2. Gap efficiency — avoids creating orphan gaps too short to be useful.
3. Day spread — prefers days with fewer meetings to distribute load.
4. Buffer comfort — rewards slots with breathing room before and after.
5. Booking density — penalizes already-packed days.

The result: your AI assistant picks the same slot you would. Not the first available time — the best available time.

The dry-run safety net

Every booking action runs in dry-run mode by default. The assistant shows you exactly what it plans to do:

• The selected time slot and its quality score
• The guest name and email
• The event and duration
• What confirmation emails will be sent

You review and confirm. Or you say "try Wednesday instead" and the assistant adjusts. Full control, zero friction.

Once you're comfortable, you can tell the assistant to book directly — skipping the preview for routine meetings while keeping it for high-stakes ones.

Why this isn't just "another integration"

Calendar integrations have existed for years. Zapier can create a booking when a form is submitted. Your CRM can trigger a scheduling link. But those are automations — rigid, trigger-based, no intelligence.

MCP is different because it gives the agent judgment. The agent doesn't just execute a script. It understands context ("mornings preferred"), evaluates trade-offs (gap efficiency vs. day spread), and makes recommendations. It's the difference between a vending machine and a concierge.

When we built the MCP server, we realized the protocol isn't an add-on to the product. It's a second front door — one designed for the fastest-growing class of users: AI agents.

Getting started

If you have a skdul account and use Claude or ChatGPT, you're two minutes away from never manually scheduling a meeting again. The MCP server is free, included on every plan, and works with any MCP-compatible client.

For the full technical setup including troubleshooting, see our complete MCP connection guide. For the bigger picture on where autonomous scheduling is heading, that's worth reading too.

Your AI assistant already handles your email, your research, and your data. It's time it handled your calendar.