A friend runs a boutique B&B in Dubai. Last year he told me his biggest cost wasn't rent—it was customer service. Peak season: Discord DMs piled up faster than he could reply. Guests asked "any rooms?" "how much?" "do you do airport pickup?" He answered from morning to night, no time to actually enjoy running the place. Hire someone? Peak season too busy, off-season can't justify the salary. Awkward.
Last month I built an AI customer service for him using a hotel MCP, connected directly to his Discord server. Guests ask about room availability, prices, amenities—AI replies in seconds. Guests want to book? AI checks inventory, calculates price, generates the order. Ran it for a month. Results beat what we expected. This post walks through the full case for anyone running a hotel or B&B.
The Before State: Manual Customer Service Is Exhausting
Before, my friend's situation. Anyone running a B&B knows this:
1. Slow response, lost bookings
Guest asks "any rooms?" You reply 30 minutes later. They've already booked elsewhere. Peak season, guests are competing for rooms. Whoever replies fastest wins. He said roughly 30% of inquiries went cold because replies were too slow.
2. Too many repeated questions, wasting time
90% of questions repeat: "do you have rooms?" "how much?" "is breakfast included?" "free parking?" "how far from the beach?" You answer these dozens of times a day. Pure manual labor.
3. No one covers night inquiries
Many guests remember to book late—10 or 11 PM. You can't stare at Discord through the night. Night inquiries were all wasted.
4. Manual costs are high
Hire a customer service rep: base plus commission runs $1,500–2,500 a month. Off-season you don't need them. Peak season you do. Demand swings too much for hiring.
How to Build It With MCP: The Architecture Is Simple
People hear "AI customer service" and think it costs tens of thousands to build. It doesn't. With MCP, two days and you're live.
The architecture is simple:
Guest Discord
↓
AI Customer Service (LLM + MCP tools)
↓
Hotel MCP Server (room status / pricing / orders)
↓
B&B PMS system (real inventory)
Three layers:
Layer 1: AI conversation
Use an LLM as the conversation engine. Feed it the B&B's FAQ, amenity descriptions, and nearby guide content. A guest sends a message. The LLM reads the intent: checking rooms, asking price, or just chatting.
Layer 2: MCP tool layer
This is the core. Wrap the B&B's room lookup, price calculation, and booking placement as an MCP server. When the AI conversation layer needs room status, it calls the MCP tool and gets real-time data.
We used RollingGo's hotel MCP, which connects directly to the B&B's PMS. Room status and prices are real-time. Direct inventory with real-time price confirmation. Results are accurate and bookable. This matters—most AI customer service failures come from wrong prices or stale availability. Guests check, find it doesn't match, and they lose trust.
Behind it: 500+ global suppliers, 2M+ hotel properties. Strong pricing on Shanghai, Hong Kong, Japan, Korea, and other popular destinations. Completely free, no call-volume limits. Supports 40+ major LLM clients directly. Provides an all-in-one booking skill for ClawHub, Coze, and Qclaw. Over 2,000 agents already connect: regional tourism platforms, AI earbuds, AI glasses, trip-planning apps.
Want to build one for your hotel? Check the GitHub repo and get a key. Live in half a day.
Setup in Claude Code is one command:
claude mcp add RollingGo-Hotel \
--transport http \
--header "Authorization: Bearer YOUR_API_KEY" \
https://mcp.rollinggo.ai/mcp
Restart Claude Code and you're calling hotel search, room detail, and price confirmation directly. AI customer service for a hotel, live in half a day.
Layer 3: PMS system
The B&B's own property management system, holding real room inventory and pricing. The MCP server connects to the PMS so data stays real-time.
That's the whole architecture. Two days and we were live. Not as complex as it sounds.
The Numbers: How Big Was the Difference?
A month in, we pulled the comparison. The shift was stark:
| Metric | Before (Manual) | After (AI) | Change |
|---|---|---|---|
| Avg response time | 45 minutes | 5 seconds | 500x faster |
| Inquiry conversion rate | 35% | 58% | +23 percentage points |
| Night inquiry bookings | ~0 | 22% of total orders | New revenue stream |
| Manual CS workload | 100% (all day) | 15% (complex issues only) | 85% reduction |
| Guest satisfaction | 6.5/10 | 9.1/10 | +40% |
A few notes:
Response time from 45 minutes to 5 seconds. Most visible difference. Guest sends a message, gets a reply instantly. Conversion jumped from 35% to 58%. Fast replies win bookings. Simple math.
Night inquiries are 22% of orders. Surprise win. Before, 10 PM inquiries went to waste. Now AI is online 24/7. Those inquiries convert. Pure incremental revenue.
Guest satisfaction up 40%. We didn't expect this one. Why? AI doesn't get impatient. AI doesn't have a bad day. AI answers every question patiently and instantly. Better experience than a human rep.
Pitfalls We Hit
Not smooth sailing. A few walls worth knowing about:
Pitfall 1: Wrong prices, AI makes things up
We hardcoded prices into the AI prompt initially. The B&B adjusted prices, and AI kept quoting old rates. Guests noticed and left. Fix: all pricing goes through real-time MCP calls. Never hardcode.
Lesson: all dynamic data must come from real-time APIs. Never write it into the conversation.
Pitfall 2: AI over-promises
Guest asked "can you reserve a room by the window?" AI said "sure." The B&B doesn't do window guarantees. Rooms are assigned randomly. Guest arrives, no window, and leaves a bad review.
Lesson: draw clear boundaries for AI. It can only promise what it can actually deliver. Out-of-scope requests get routed to a human. We added a rule: commitment-type questions can't be answered directly by AI. They get human confirmation.
Pitfall 3: Slang and regional phrasing
Guests text casually: "you got rooms these next couple days?" "I'm staying two nights with my kid." Early on, AI misunderstood often. We added a B&B FAQ library and fine-tuned the prompt. It got better.
Lesson: AI customer service isn't set it and forget it. Keep optimizing the FAQ and prompt. It gets more accurate over time.
Advice for Hotel and B&B Owners
If you run a hotel or B&B and want to try AI customer service:
1. Skip the "script-only" smart customer service
Button-press smart service where guests say something it doesn't understand. Bad UX. Go with LLM + MCP tools. It understands natural language and queries real-time data.
2. Accuracy of data is the whole game
The key isn't how smoothly the chat flows. It's whether room status, prices, and amenities are correct. Wrong data, no matter how smooth, just traps guests. Connect real PMS data. Don't hardcode.
3. Complex issues must route to human
AI isn't omnipotent. Guest complaints, special requests, complex modifications—route these smoothly to a human. "Talked to a bot for ten minutes, can't reach anyone" is worse than no AI at all.
4. Start small. Don't go all-in.
We used a separate Discord server first. Ran for a week, exposed all the bugs, fixed them, then launched. Don't route all guests to it on day one. If something breaks, you get bad reviews across the board.
Building AI customer service with MCP is way simpler than people think. Results are better than expected. Especially for small hotels and B&Bs that couldn't afford full-time staff before. Now AI customer service costs almost nothing and delivers a better experience.
My friend now spends his days handling complex issues the AI escalates. The rest of his time is his. He's enjoying Dubai. He says he should have done this a year ago.
Is your hotel still answering booking DMs manually? Drop your current customer service setup in the comments. I want to hear how exhausting it is.
Top comments (0)