Smily MCP: User Guide

Source: Smily webinar recordings, September 2026, and technical documentation available here.

1. What is Smily MCP?

Smily MCP connects an AI assistant (Claude, ChatGPT, etc.) directly to your live Smily account data via the Model Context Protocol (MCP). Instead of exporting data or waiting on custom dashboards, you ask the AI questions in plain language and it queries your real data to answer, build reports, and flag issues.

Access is read-only. The AI can query data but cannot create, modify, or delete anything in your Smily account.

Smily currently exposes 70+ data "resources" through MCP and more are being added. Examples: bookings, daily rates & rate rules, reviews, tasks, revenue, protection programs (damage & cancellation protection), messaging/inbox. According to Smily, the MCP already exposes more data than Smily's own public API.

Smily's stated priority for this feature is data quality: the underlying figures are only useful if they're clean and correctly computed. This is presented as one of Smily's core strengths versus other platforms in the market.

2. Which AI assistant should you use?

• Claude is Smily's recommendation. Claude currently gives the best intelligence/performance for this use case (not the cheapest option, but the most capable).
• ChatGPT is also supported and works the same way, just with a different setup flow (below).
• Claude Cowork / Claude Code additionally lets the AI act (schedule recurring reports, refresh dashboards on a timer). Claude Chat can only analyze and answer, not schedule.

3. Prerequisites

You need a personal MCP API token (a private key, treated like a password) to authenticate.

• Not yet available directly in the Smily interface.
• Request it from your Account Manager or Smily Customer Support.
• Keep it private. It grants read access to your account data.

4. Connecting to Claude (Desktop / Cowork)

1. Use the Claude Desktop app, open Cowork.
2. Customize connectors → + Add a custom connector.
3. Name: "Smily".

4. URL (same as above, legacy domain):

   <https://www.bookingsync.com/mcp?token=YOUR_PRIVATE_TOKEN>

5. Click Add. You now have read-only access to your Smily account data inside Claude.

5. Connecting to ChatGPT

1. In ChatGPT: profile (bottom left) → Settings → Apps → Create app.
2. Icon: optionally use the Smily logo.
3. Name: something you'll recognize easily, e.g. "Smily – [your account name]", especially useful if you have several apps/accounts connected.
4. Description: note that it's read-only, so there's no risk of erasing or changing anything in Smily.

5. Connection URL:
   (Smily's API still runs on the legacy "BookingSync" domain.)

   <https://www.bookingsync.com/mcp?token=YOUR_PRIVATE_TOKEN>

6. Authentication: select "no authentication" — the token in the URL is the authentication.
7. Confirm that you understand this app isn't built by OpenAI, then click Create.
8. Done! You can now chat with your Smily data in ChatGPT.

6. Example Use Cases & Prompts

6.1 Operational task analysis

"Analyze my operational tasks for the last two months: which task categories are most reported, which properties come up most often, and where actual duration significantly exceeds planned duration."

Note: this only works well if your team actually logs task data (planned time, actual time, completion status) in Smily. If a task was never marked complete or timed, the AI can't invent that data, it will tell you so rather than guess.

6.2 Guest communication / inbox analysis

"Over the last 90 days, evaluate my traveler communication: average response time and overdue conversations. Over the last 12 months, identify the topics/questions that occur most often in traveler messages."

Useful for spotting SLA problems (e.g., multi-day response times) and recurring questions you could pre-empt with automated notifications (e.g., "how do I check in?" sent automatically a week and two days before arrival).

6.3 Pricing / length-of-stay (LOS) audit

"Do a complete pricing audit for [property]. Explore the length-of-stay discounts, seasonal rates, minimum-stay rules, and active special offers. Flag anomalies: nights with no rate set, abnormal gaps between neighboring periods, inconsistencies."

You don't need to specify exactly what to look for, a broad "flag everything, take your time" prompt works. The AI may take 30–60 minutes to process; you don't need to sit and watch it.
Important: this is a complement to a revenue manager, not a replacement. The AI doesn't have full market context, but it's very good at catching data inconsistencies at scale.

6.4 Account health check (dashboard)

"Use Smily and give me a complete account health check in HTML: occupancy and revenue over the last 12 months vs. the previous 24 months, average review trend, share of direct bookings vs. platforms, share of properties enrolled in Smily Protect and in cancellation protection."

Good for a founder/CEO-level view of the business. This can also be built with financial figures automatically blurred, so you can demo it to prospective owners without exposing other clients' real numbers.

6.5 Owner report

"Use Smily and write an owner report for [Property Name]: revenue and occupancy this year vs. the previous two years, what reviews say about the property's condition and amenities, recurring maintenance issues reported in tasks, and 2–3 recommendations to preserve the property's value and improve long-term profitability. Make it HTML with [yourdomain].com's branding, including the logo."

The AI can pull your logo/colors directly from your website, usually needs a correction pass or two to get the exact shade/branding right. It can also build a maintenance issue timeline straight from review text (e.g., "bad access road flagged July 2024… still flagged May 2025… confirmed fixed May 2026"), which is otherwise very time-consuming to compile manually.

6.6 Weekly KPI one-pager (reusable, schedulable)

"Build a one-page HTML report in [yourdomain].com's colors and logo, with this week's KPIs as charts: occupancy, revenue vs. previous week, revenue vs. last season, overdue tasks, share of direct bookings. Give me the top 3 actions for this week. Save it as a reusable template / artifact so I can relaunch it every week."

In Claude Cowork, once you have this as an artifact, you can literally say "refresh this every Monday at 9:00 a.m." and it will re-pull fresh data on schedule without you re-prompting. Note this consumes more AI usage/credits, scheduled recurring refreshes may require a higher Claude plan.

6.7 Email notification analytics

"Study all email notifications sent over [period]. Build a dashboard: open rate, click rate, bounce rate. Tell me what's working, what isn't, and give me top recommended actions."

Useful for retargeting/re-engagement performance (a factor in direct-booking share). Note: at the time of the webinar, open-rate/click-rate data wasn't fully wired through the MCP yet.

7. Working Well With the AI: Practical Tips

• Verify computed figures, especially financial ones. In the demos, a "% direct bookings" figure came out wrong because the AI only counted a subset of channels as "direct." Ask it to explain its method — most reports include a short recap of what was computed and how, which makes it easy to sanity-check.
• Iterate. Branding, colors, and scope rarely land perfectly on the first try. Correct it conversationally ("that's not quite the right blue, please recheck the site") — the more you correct it within a session, the more accurate it gets.
• Be specific about goals — precise KPIs, exact comparison periods, exact branding source — for sharper output. Vague prompts give generic answers.
• Voice input speeds things up. Smily's own demos were done using a dictation tool (Whisper Flow) rather than typing long prompts.
• You can convert reports to PDF by asking the AI directly, or via your browser's Print → Save as PDF.
• You can ask the AI to draft the owner email that goes with a report, and even schedule a follow-up ("if no reply in 3 days, send a short follow-up"). See the security note below before automating sending.

8. Security & Data Handling

• The MCP connection itself is read-only, there is no risk of the AI overwriting or deleting anything in your Smily account through this connector.
• Be cautious about where you share or host reports. These reports often contain confidential owner data, including financial metrics. Avoid publishing them somewhere public (e.g., directly on a public website); prefer direct channels (email, WhatsApp) or access-controlled portals.
• If you connect the AI to your email system to send messages automatically, understand that this carries its own security risk (an AI agent with send access to your inbox). Review that risk before enabling automated sending.
• AI can be wrong about numbers, especially financial ones, this was called out explicitly in the webinar as a real risk. Always keep an eye on the order of magnitude and verify anything owner- or investor-facing before sending it out.
• If you go beyond read-only MCP queries and start building custom tools that write to a database (e.g., "vibe-coded" internal tools), this requires real engineering practices: version control (e.g., GitHub), and verified database backup/replication. Without that, a bad AI-assisted change can wipe or corrupt data. Smily's own infrastructure, for reference, is fully replicated with backups retained for months. That level of care is recommended if you're building your own tools on top of the data.