Your Favorite Software Has Gaps. MCP Fills Them.

How I used Claude and a Model Context Protocol to solve a problem my property management software couldn’t — and why this pattern matters for every SaaS tool you use.

Prefer to watch? I walked through this 2 minute Loom video — it’s a summarized version. The companion slides go deeper than the video. The full story is below.

My wife and I run a small Airbnb in my backyard in Ruston, Louisiana. It’s a guest favorite and my wife has been a superhost since 2021.

The reason I’m writing this: I recently closed a gap in our property management software in about 20 minutes, without writing a line of code, using an MCP server. What I got back wasn’t really the feature — it was peace of mind. I’m no longer the middleware of worry between two systems, tracking what might fall through the cracks. That trade alone is worth the whole article.

We use Hospitable as our property management system. It’s genuinely excellent software. Automated messaging, calendar sync across different STR platforms, task management, direct booking — it handles the heavy lifting of running a short-term rental without requiring us to babysit it constantly.

But it has a gap, like any software.

The Gap

During the summer, we don’t want same-day turnovers. When a guest checks out at 10 AM, we need that day for cleaning and prep before the next guest checks in at 3 PM. We can normally swing that, but we are traveling this summer for 6 weeks, and need at least a day for turnover. My mother in love will be our POC for clients and I want her to have this extra margin.

Hospitable has a turnover time setting — but it’s all or nothing. You either enforce a buffer between every reservation year-round, or you don’t. There’s no way to say “block same-day turnovers only between July 7 and August 14.”

This is a reasonable product decision. Most hosts either want a buffer or they don’t. Building date-range-specific turnover policies adds complexity for a niche use case. I get it.

But I still needed the feature.

The Old Way

Before MCP, my options were:

1. Manual calendar management. Every time a new booking came in during the summer window, I’d check the checkout date and manually block that day from check-in. This works until you forget — and you will forget, because it depends on a human remembering to do a thing every time a booking arrives.
2. Set the buffer year-round. This costs occupancy during months when same-day turnovers are fine. For a property that generates roughly $10K/year before fees, every lost night matters.
3. Live with it. Accept the occasional stressful same-day turnover and hope it doesn’t tank a review, stress out my mother in love, etc.

None of these are great. The first doesn’t scale, the second costs money, and the third is a gamble.

The MCP Way

Model Context Protocol is an open standard that lets AI agents interact with software through structured tool calls. Think of it as giving Claude (or any AI) the ability to press buttons in your software — not by clicking around a screen, but by talking directly to the application’s API through a standardized interface.

Hospitable has an MCP server. So does my calendar, my email, my iMessage, and dozens of other tools.

Setting one up isn’t a developer project. Adding Hospitable to Claude took me about a minute: click connect, an OAuth window pops up in the browser, I authorize, and Claude can see my reservations. No API key to generate, no credentials to copy-paste, no integration code to write. If you’ve ever signed into a website with “Continue with Google,” you already know how this feels.

Here’s what I built in about 20 minutes with Claude:

1. Fetch reservations from Hospitable with checkout dates in my summer window (July 7 – August 14)
2. Filter to only newly booked reservations (so it doesn’t reprocess ones already handled)
3. Check the calendar to see if the checkout date is already blocked for check-in
4. Block check-in on checkout dates that aren’t already blocked
5. Handle cancellations — if a reservation gets cancelled, check whether the block is still needed (another reservation might share that checkout date), and unblock if it’s safe
6. Notify Ashley via iMessage if any changes were made

The whole thing runs as a scheduled task, once a day at 8 AM. No human in the loop. No forgetting. No spreadsheet to maintain.

The first time I ran it, Claude pulled my reservations through the Hospitable MCP server, identified four in the window, blocked check-in on each checkout date, and confirmed the calendar updates. It took about 30 seconds.

This Isn’t About Airbnb

The pattern here is bigger than short-term rentals. Every piece of software you use has gaps — features that almost do what you need, but not quite. Settings that are too coarse. Workflows that require manual steps between two systems.

Consider these analogies:

CRM that can’t do conditional follow-ups. Your CRM sends automated sequences, but you want to pause a sequence if the prospect opened a specific email and didn’t click — and then route to a different cadence. The CRM has sequences and it has analytics, but the two don’t talk to each other the way you need.

Project management with rigid automation. Your PM tool can auto-assign tasks based on status changes, but you need it to also check the assignee’s calendar and redistribute if they’re on PTO. The PM tool and the calendar tool both have APIs, but nobody’s connected them for your specific workflow.

Accounting software with inflexible categorization. Your bookkeeping tool auto-categorizes transactions, but your business has edge cases — a single vendor that supplies both office supplies and equipment, and you need the categorization to depend on the invoice amount. The rules engine can’t express that condition.

In every case, the software is good. The gap is small. But the gap is exactly where manual work, errors, and inconsistency creep in.

MCP turns those gaps into solvable problems. Instead of waiting for the vendor to build the feature, or hiring a developer to write a custom integration, you describe the logic to an AI agent that already speaks the language of both systems.

Why This Matters Now

Three things are converging:

MCP adoption is accelerating. More software vendors are shipping MCP servers because they recognize that AI interoperability is becoming table stakes. Hospitable, Slack, Google Calendar, Gmail, HubSpot, Jira — the list grows weekly.

AI agents are getting reliable enough. A year ago, I wouldn’t have trusted an automated agent to modify my Airbnb calendar unsupervised. Today, with structured tool calls, idempotent operations, and clear guardrails, the reliability is there. My skill checks the calendar before making changes, cross-references accepted and cancelled reservations, and only messages Ashley when something actually changed.

The economics favor automation. The Cozy Cottage is a $10K/year side business. I can’t justify hiring a developer to build a custom Hospitable integration. But I can spend 20 minutes describing the logic to Claude and have a working, scheduled solution that runs indefinitely.

Always UAT — Even When the Agent Says It’s Done

I was feeling pretty good about all of this — until my wife Ashley looked at the calendar and said: “That’s awesome. But what about the front end? You blocked check-in on checkout days, but you didn’t block checkout on days where we have an existing check-in.”

She was right. I’d only handled one direction. If a guest was checking in on a Saturday, nothing prevented a different guest’s reservation from ending that same Saturday morning. Same-day turnover, just from the other side.

The fix was straightforward. I opened a new Cowork session, edited the skill to block in both directions (check-in on checkout dates and checkout on check-in dates), and tested until the calendar showed restrictions on both ends of every reservation.

Here’s the part that made me smile: the scheduled task didn’t need to change at all. The prompt I’d written was simply “read the skill file and follow it.” The skill file is the source of truth. Update the file, and every future run picks up the new logic automatically. No redeployment. No reconfiguration. Just edit the instructions and the agent adapts.

The incremental operating cost of all of this? Zero. My Hospitable plan already includes MCP access. Claude Cowork starts at $20/month. The automation itself costs nothing to run.

A Note on Trust

Reasonable question if you’re new to this: should I be worried about an AI agent operating my software?

A few guardrails I follow that should give you some comfort.

Stick to first-party MCP servers. I use the MCP server published by Hospitable — the company that builds my property management software — and MCP integrations published by Anthropic, the company that builds Claude. I don’t install random third-party MCP servers from the internet. The trust model is simple: the people who build the software, or the people who build the AI, are the right people to write the bridge between them. Anyone else is asking for trust they haven’t earned.

Claude asks before it acts. The first time Claude tries to use an MCP tool, it pauses and asks me. I get three choices: deny it, approve just this once, or approve for future runs. Nothing happens silently. I read what it’s about to do and decide.

Drafts, not sends. For anything sensitive — outbound messages, calendar invites, anything client-facing — I tell Claude to draft, not send. It does 95% of the work: writes the message, attaches the right context, formats it correctly. I review the last 5% and hit send myself. That’s the right division of labor, especially early on while you’re building trust with the agent.

Pro tip for scheduled tasks. If you’re scheduling something to run unattended (like my 8 AM daily run), trigger it manually once right after you create it. Claude will prompt you to approve the tool calls, and once you approve “for future runs,” the scheduled run won’t get stuck waiting for permission later. Save yourself the head-scratching when you find out the next morning that nothing fired.

The Mindset Shift

The real unlock isn’t technical — it’s a shift in how you think about software limitations.

The old mindset: “This software doesn’t support what I need. I’ll submit a feature request, find a workaround, or switch products.”

The new mindset: “This software doesn’t support what I need. I’ll describe the gap to an AI agent that can operate the software on my behalf.”

The gap between what software does and what you need it to do is where MCP lives. It’s not replacing your tools — it’s making them work the way you actually need them to.

What to Try First

If you want to feel this for yourself, don’t start with a big project. Pick one tool you use every day, check if it has an MCP server, and ask Claude to do the next thing on your to-do list inside that tool. Watch it work. Then ask, “what’s the next gap?”

Many of the SaaS products you already pay for have MCP servers ready to go. A non-exhaustive list of popular ones: Notion, Slack, GitHub, Stripe, Sentry, HubSpot, PayPal, Asana, Linear, Atlassian — plus, for the finance-and-ops crowd, QuickBooks, Sage, and NetSuite. (Anthropic maintains a directory, though it isn’t exhaustive — Hospitable, for example, isn’t on Anthropic’s list yet but has its own MCP server, which is exactly why I’m writing this. Check your vendor’s docs.)

Want help applying AI and MCP to your business workflows? McCandless Consulting offers AI enablement training and strategy for teams that want to close their own software gaps.

P.S. — If you’re a host (or thinking about becoming one). My wife and I have coached four other hosts into becoming superhosts themselves. We do it for free — Airbnb gives us a referral incentive, and all four of our referees have become superhosts and/or hosts of guest favorites. If you want to chat about hosting, two pieces I’ve written:

• People ask me about running an Airbnb… — what I tell folks who ask whether they should try it
• How Airbnb hosts can analyze their listings with Tableau — for the data-curious hosts