How to Get Your Dealership MCP-Ready: A 90-Day Playbook

The last four dispatches made the case: MCP is the port, the point-to-point era is ending, and the neutral layer wins. Fine. But Monday morning you're staring at a used-car aging report, not an architecture diagram. What do you actually do?

This is the do-something dispatch. Ninety days, three phases, no code, no consultants required. Everything here is work your existing team can execute — and every bit of it pays off even if you never touch an AI agent, because it's the same hygiene that makes every other initiative in the store run better.

PHASE 1 · DAYS 1–30Know what you're actually running

Most stores cannot produce a complete list of their own software. Not because anyone's careless — because the stack accreted over fifteen years across three GMs and nobody owns the map. Make the map.

The stack audit

One spreadsheet, one row per system: what it is, what it costs, who owns the contract, when it renews, what data lives in it, and — the column nobody has — how data gets out. API? Nightly file drop? A report someone exports by hand? Nothing at all?

The two vendor questions

Email every vendor on the list this month — before your renewal, not during it:

1. "Do you support MCP today, and if not, when?" A roadmap answer is acceptable. A blank stare is data too.
2. "What does programmatic access to our data cost?" If the answer to reaching your own customer and inventory data is a four-figure integration fee, you've found a toll booth — and a renewal negotiation topic. (See the integrations dispatch for why this leverage is shifting to you.)

PHASE 2 · DAYS 31–60Clean the data, write the rules

An AI agent is a merciless auditor. Point one at your CRM and it will find the four records for the same customer, the "test" leads from 2021, and the trade-in valued at $4. Fix the worst of it before an agent embarrasses you in front of a customer.

Data hygiene, prioritized

• CRM dedupe first. Agents personalize from customer history; fragmented records mean an agent greeting a 10-year customer like a stranger.
• Inventory truth second. Photos current, statuses real, options accurate. An agent will happily quote a car that sold last Tuesday if the feed says it's in stock.
• Service records third. Last-service dates and declined-work history are the fuel for the highest-ROI agent use case: bringing customers back.

The one-page governance policy

Write it now, while it's theoretical and nobody's defensive. Four sections, one page:

This maps one-to-one onto how a proper MCP layer enforces permissions — write the policy now and it becomes configuration later, as described in the universal server dispatch. And while you're at it: kill the shared logins. Every one of them is a hole in any future audit trail.

PHASE 3 · DAYS 61–90Pilot one workflow. One.

The failure mode here is ambition — trying to "do AI" across the whole store and drowning. Pick a single workflow that's narrow, measurable, and annoying enough that the team wants it automated.

Best first pilots

• After-hours inventory questions. Highest volume, lowest risk: the 9pm "do you have a Tacoma under $35k" text, answered from live inventory in seconds. (The worked example in TX-001.)
• Service appointment booking. Bounded, measurable, and customers prefer texting anyway.
• Declined-work follow-up. The money printer nobody staffs: customers who declined brakes six months ago, contacted with a specific, permissioned offer.

Run it like a pilot, not a leap of faith

1. Human-in-the-loop for the first two weeks: the agent drafts, a person approves. Loosen only what earns it.
2. Measure three numbers: response time, appointments set, and escalations to humans. Ignore vanity metrics.
3. Review the audit log weekly with the team — the fastest way to build trust is showing exactly what the agent did and didn't do.
4. At day 90, decide with data: scale it, fix it, or kill it. All three are wins compared to guessing.

What you'll have on day 91

A map of your stack. Vendors on record about MCP and data access. Cleaner data than you've had in years. A governance policy that took an hour instead of a crisis. And one working, measured AI workflow — while most of your market is still forwarding each other think pieces.

That's the position from which everything in the adoption timeline gets easier: you'll evaluate vendor MCP claims from experience, not hope, and every new agent you add plugs into rails you've already laid.