MCP for Hospitality: AI Agents and Your CRM
Model Context Protocol — usually shortened to MCP — is the open standard that lets AI agents like Claude and ChatGPT connect to external systems. For hospitality, MCP is the bridge between general-purpose AI and the operational data living in your CRM. It is the difference between asking Claude a generic question and asking Claude a question grounded in your actual guests, reservations, and revenue.
Key Takeaways: Model Context Protocol (MCP) is an open standard introduced by Anthropic that lets large language models talk to external systems through a common interface. SendSquared exposes an MCP server so external AI agents — Claude, ChatGPT, custom copilots — can query and act on CRM data with proper authentication and scoped permissions. Practical use cases today include natural-language CRM queries, campaign draft generation grounded in real data, and internal copilots for ops and revenue teams. MCP is early but compounding fast.
What MCP Actually Is
MCP is a protocol — like HTTP or USB-C — that defines how AI agents request information and trigger actions in external systems. Anthropic introduced it in late 2024 and it became an industry standard within twelve months. By 2026, every major AI assistant supports MCP and most major SaaS products either ship an MCP server or are building one.
The protocol has three core ideas.
Servers expose data and actions. A SaaS product like SendSquared runs an MCP server that exposes specific tools and resources. The server defines what the AI can read (guest profiles, reservations, campaign performance) and what it can do (draft a campaign, segment a list, send a message after human approval).
Clients consume the protocol. Claude Desktop, ChatGPT, custom-built agents, and IDEs all support MCP as clients. Once a user connects a client to a server, the AI can use the exposed tools.
Authentication is built in. MCP supports OAuth and scoped tokens. The agent only sees what the authenticated user has permission to see. The same role-based access controls that govern the web app govern the MCP interface.
The result is that an AI agent becomes a thin layer on top of the systems the user already uses, instead of a parallel universe with its own incomplete data.
Why Hospitality Should Care
Hospitality data is high-volume, high-context, and operational. Guests have stay histories. Reservations have channels. Properties have rates and seasons. Survey responses have sentiment. None of this is useful to an AI agent unless the agent can actually access it.
Without MCP, an AI assistant is generic. It can write a generic pre-arrival email. It cannot write a personalized one because it does not know the guest.
With MCP, the same AI assistant becomes specific. It can pull the actual guest, the actual reservation, the actual stay history, and produce something grounded in real data. It can answer “how many promoters do we have in our Florida market who have not stayed in 12 months” because it can query the CRM live.
The SendSquared external AI agents integration exposes the CRM data model through MCP so any compatible AI client can use it.
Practical Use Cases Today
The use cases that work in 2026, in roughly the order operators adopt them.
1. Natural-language CRM queries. “How many guests in the Asheville market have lifetime value above $5K and have not stayed in 18 months?” Claude returns the count, the guest segmentation definition, and offers to build a campaign for the segment. No SQL, no segment builder UI, just the question.
2. Survey response summarization. “Summarize the last 30 days of detractor responses by theme.” Claude pulls the responses, clusters them, and produces a digest with examples. The GM gets actionable insight in 30 seconds instead of 30 minutes.
3. Campaign draft generation grounded in real guest data. “Draft a win-back campaign for our top-tier repeat guests targeting fall stays.” Claude pulls the segment, references prior stay patterns, and drafts subject lines, body copy, and SMS variants. Human approval before send.
4. Custom internal copilots. Mid-size operators build internal copilots — a “revenue copilot” or an “ops copilot” — that wrap MCP access to SendSquared (and to the PMS, the channel manager, the BI tool) with custom prompts and guardrails. The team uses the copilot like a colleague.
5. Daily briefings. A scheduled query runs each morning: “What detractor responses came in overnight? What VIP arrivals are today? What gaps are open in the next 7 nights?” Claude produces the briefing and delivers it to Slack or email.
These use cases compound. The same MCP connection that powers daily briefings also powers ad-hoc queries and campaign drafting.
How It Works in Practice
The user flow is straightforward.
One-time setup. The user opens Claude Desktop (or another MCP-compatible client), navigates to the integrations panel, and authorizes the SendSquared MCP server with OAuth. The server registers with the client and exposes its tools.
Ongoing use. The user types a question in Claude. Claude decides whether to use the SendSquared tools. If yes, Claude calls the appropriate tool, receives the data, and produces a response grounded in that data.
Permissions. The user only sees what their CRM role allows them to see. A property-level user does not see other properties. A reporting-only user cannot trigger sends.
Audit trail. Every action taken through MCP is logged in the CRM audit trail with the user identity and the agent identifier.
The technical surface is invisible to most end users. The experience is “I asked Claude a question about my CRM and Claude answered.”
Why MCP Beats Custom Integrations
Before MCP, every AI integration was bespoke. Every SaaS product built its own ChatGPT plugin, its own Claude integration, its own enterprise AI connector. The result was duplicated work and limited interoperability.
MCP standardizes the interface. The same MCP server that powers Claude Desktop also powers ChatGPT, custom internal agents, and any future MCP-compatible client. Build the server once, work everywhere.
For SendSquared specifically, this means the AI voice layer, the unified inbox, the marketing automation engine, and the survey platform all expose their data through one consistent protocol. External AI agents pull from one source of truth.
What MCP Does Not Do (Yet)
MCP is early. A few things to know before assuming it solves everything.
It is not autonomous. MCP gives agents access to data and tools. It does not give them permission to act unilaterally. Most workflows still require human approval before send.
It does not replace the CRM UI. Day-to-day operational work still happens in the CRM. MCP is the layer for ad-hoc queries, summarization, drafting, and copiloted analysis.
Quality varies by client. Claude is currently the strongest MCP client. ChatGPT support is improving. Open-source clients are uneven. The experience depends on the client.
Security boundaries matter. Always use scoped OAuth tokens. Never expose full admin access through MCP. The same security hygiene that applies to API keys applies here.
What to Do This Quarter
If your CRM exposes an MCP server (SendSquared does), the on-ramp is short.
- Connect Claude Desktop or ChatGPT to your CRM via MCP. 10 minutes.
- Run five natural-language queries you would normally build a report for. See what comes back.
- Draft a campaign with Claude grounded in real segment data. Review and refine.
- Set up one daily briefing query relevant to your role.
- Share the workflow with one teammate and watch them adopt it within a week.
By the end of a quarter, MCP-driven workflows are part of the team’s daily rhythm. The CRM data becomes more useful because the access cost dropped to zero.
The Bottom Line
Model Context Protocol is the open standard that connects AI agents to external systems. For hospitality, MCP turns generic AI into specific AI grounded in real CRM data. The use cases today — natural-language queries, campaign drafting, response summarization, daily briefings, custom copilots — are practical and compounding.
SendSquared’s MCP server exposes the full CRM data model with OAuth authentication and role-based scoping. Connect any MCP-compatible client and the team gets a copilot that actually knows your guests.
Want to see MCP running on real hospitality data? Explore the external AI agents integration or book a demo →
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.
Frequently Asked Questions
What is the Model Context Protocol?
MCP is an open standard, originally introduced by Anthropic, that lets large language models like Claude and ChatGPT connect to external systems through a common interface. Think of it as USB-C for AI agents.
How does MCP relate to hospitality?
It lets external AI agents query and act on your CRM data. A property manager can ask Claude in natural language for guest insights, segmentation, or campaign drafts and the response is grounded in real CRM data.
Is MCP secure?
MCP supports authentication and scoped permissions. SendSquared's MCP server uses OAuth and respects the same role-based access controls as the web app. The agent only sees what the user has permission to see.
Do I need to be a developer to use this?
No. End users connect Claude or another MCP-compatible client to SendSquared once and then ask questions in plain English. Developers build custom copilots on the same protocol when they need deeper workflows.
What can I actually do with it today?
Natural-language CRM queries (lifetime value lookups, segment counts, recent feedback summaries), campaign draft generation grounded in real guest data, and custom internal copilots for ops and revenue teams.