Using AI Agents (Claude, ChatGPT) With Your Hotel CRM

Connecting an AI agent like Claude or ChatGPT to your hotel CRM is no longer a research project. In 2026, it is a 30-minute setup that turns a generic assistant into a colleague who knows your guests, your reservations, and your revenue. The glue is Model Context Protocol — MCP — and the result is a CRM you can talk to.

Key Takeaways: AI agents like Claude and ChatGPT can now query and act on your hotel CRM through Model Context Protocol (MCP), an open standard that exposes data and tools to any compatible AI client. Practical use cases for operators today include natural-language queries against live reservation data, drafting personalized campaign briefs, summarizing reviews and survey responses, and surfacing anomalies before they hit revenue. SendSquared’s external AI agents MCP server makes this work with tenant-scoped tokens, role-based permissions, and full audit trail — so the same security model that governs your CRM web app governs the AI layer.

---

What MCP Is, In One Paragraph

Model Context Protocol is an open standard introduced by Anthropic in late 2024 that defines how AI agents request information from and trigger actions in external systems. Think of it as USB-C for AI assistants. A SaaS product runs an MCP server that exposes specific tools and resources. Claude Desktop, ChatGPT, custom agents, and emerging clients across Grok, Gemini, Llama, and Mistral all consume that protocol. Connect once, and the AI can use the exposed tools the same way a developer uses an API — except the user just asks questions in plain language. For deeper background, see our explainer on Model Context Protocol for hospitality.

Why Hotel Operators Care

A hotel CRM is a data graveyard if the cost of pulling an answer is “open the dashboard, build a segment, export a CSV, and stare at it.” Most operational questions never get asked because the cost of asking is too high.

AI agents collapse that cost to zero. An ops manager who would never have logged into the CRM to check “how many guests in our Asheville market have lifetime value above $5K and have not stayed in 18 months” can now ask Claude that exact question and get an answer in 10 seconds. The answer is grounded in real CRM data, with the segment definition shown alongside.

The compounding effect is that more questions get asked. More questions get asked, more decisions get made on data, and the CRM starts paying back the investment it has been quietly making in the background.

Practical Use Cases Today

Here are the workflows operators adopt first, in roughly the order they show up.

1. Querying Reservations and Guest Profiles

The simplest and most common. “How is direct booking pacing this week compared to last?” “How many VIP arrivals do we have on Friday?” “Pull the profile for the guest who called yesterday — show me stay history and last survey response.”

Claude or ChatGPT calls the SendSquared MCP server, retrieves the live data, and produces an answer in seconds. No SQL, no dashboard navigation, no exporting. The agent is essentially a thin natural-language layer on top of the hotel CRM and the unified guest profile.

2. Drafting Campaign Briefs Grounded In Real Data

“Draft a win-back campaign for our top-tier guests in the Florida market who have not stayed in 12 months. Reference their typical booking patterns and pull example subject lines.”

The agent queries the actual segment, references prior stay data, and produces a brief with subject lines, body copy, and SMS variants. Human approval before send, but the drafting work happens in 30 seconds instead of 30 minutes. This pairs naturally with the automations layer that fires the campaign once it is approved.

3. Summarizing Reviews and Survey Responses

“Summarize the last 30 days of detractor survey responses by theme, with three example quotes per theme.”

The agent pulls the responses, clusters them, and produces a digest. A GM gets actionable themes in under a minute. This is the use case most likely to convert skeptics — the time savings are immediate and visceral.

4. Anomaly Detection

“Are there any guests this week whose support tickets have spiked above their normal pattern?” “Which properties had abandoned web sessions out of normal range yesterday?”

Anomaly detection is the workflow where AI agents punch above their weight. The agent does not need to be told what “abandoned web sessions out of normal range” means — it can reason about the data, compare to historical baselines, and flag what looks off. The operator decides what to do next.

5. Daily Briefings

Scheduled queries running each morning. “Overnight detractor responses. VIP arrivals today. Gaps open in the next 7 nights. Any agents below target on resolution rate.” The agent produces the briefing and delivers it to Slack, email, or a dashboard tile.

The same MCP connection that powers ad-hoc queries powers these scheduled briefings. Set up once, value compounds daily.

Security and Audit: What to Get Right

Giving an AI agent access to guest data is reasonable. Giving an AI agent unscoped admin access to guest data is not. The difference is in how access is gated.

Tenant-scoped tokens. Every MCP token belongs to a specific tenant — a property, a brand, or an operator. The agent cannot cross tenant boundaries. SendSquared’s MCP server enforces this at the token level.

Role-based permissions. The user who authorized the agent sees only what they have permission to see in the CRM web app. A property-level user does not see other properties. A read-only user cannot trigger sends. Same ACL, same enforcement.

Scope limits on write actions. Read-only is the default for most users. Elevated scope allows specific actions (send messages, update records, trigger automations) — and those scopes can be narrowed to “send SMS only, no campaign creation” or any combination.

Full audit trail. Every query, every action, every token use is logged with the model, agent identifier, and timestamp. The audit trail is queryable for compliance review. If a guest asks “what AI accessed my data,” you can answer with precision.

Instant revocation and rotation. Tokens can be revoked at any time and scheduled to rotate automatically. Lost laptop, departed employee, compromised account — one click and the agent loses access.

This is the same hygiene that applies to API keys. The reason it matters more for AI agents is that the surface area of “things the agent might do” is larger and the agents act faster.

Getting Started: 30 Minutes End-to-End

The on-ramp is short.

Step 1 — Generate an MCP token in SendSquared. Log in, navigate to integrations, generate a token scoped to your role. Note the server URL.

Step 2 — Connect your AI client. Open Claude Desktop, navigate to the integrations panel, paste the SendSquared server URL and token. Repeat for ChatGPT custom agents or any MCP-compatible client. Five minutes per client.

Step 3 — Run five real queries. Ask the questions you actually have. “Pull the profile for [recent guest name].” “Summarize last week’s detractor responses.” “How many leads do we have in the inquiry stage older than 7 days?” Watch what comes back.

Step 4 — Draft one campaign brief. Pick a segment you would normally build manually. Ask Claude to draft the campaign with subject lines and body copy grounded in the segment data. Approve, refine, send.

Step 5 — Schedule one daily briefing. Pick the question you most want answered every morning. Save the query. Have the agent run it on a schedule and post to Slack or email.

By the end of week one, the workflow is part of the team’s daily rhythm. The CRM has not changed. The cost of using the CRM dropped to zero.

What Not to Do

A short list of common mistakes worth avoiding.

Do not give the AI write access on day one. Run read-only for two weeks. Watch what the agent does well and where it confidently makes up details that look right but are wrong. Then graduate to scoped write access.

Do not skip the audit log review. For the first month, review the audit log weekly. Confirm the agent is doing what you expect. Tighten scopes where it is doing more than needed.

Do not assume every model is equivalent. Claude is currently the strongest MCP client. ChatGPT support is improving. Open-source clients vary. Match the model to the task — heavy reasoning on Claude, lightweight queries anywhere.

Do not replace the dashboard. MCP is for ad-hoc queries, drafting, summarization, anomaly detection. Day-to-day operational work still happens in the CRM UI. The two are complementary, not substitutes.

The Bigger Picture

Connecting AI agents to a hotel CRM via MCP is a 30-minute setup with a six-month payoff. The setup is cheap. The compounding value is in every question that gets asked because the cost of asking dropped to zero, every campaign that ships faster because the brief was drafted in 30 seconds, and every anomaly caught early because the agent flagged it before the operator would have noticed.

SendSquared was built for this. The MCP server exposes the hotel CRM, AI voice call data, hotel messaging threads, and guest loyalty signals through one consistent protocol with tenant-scoped tokens and full audit trail. Connect any MCP-compatible AI client and the team gets a copilot that actually knows your guests.

Want to see Claude or ChatGPT querying real hospitality data? Book a demo and we will run live queries against a SendSquared environment →

---

See also: external AI agents via MCP — connect Claude, ChatGPT, and any MCP-compatible AI to your SendSquared CRM with tenant-scoped tokens and full audit trail.

---

See also: hotel messaging across every channel — the unified inbox plus the messaging stack that powers it (SMS, WhatsApp, Airbnb, email, voice) with one guest profile per contact.