How True-Compass.AI built a purpose-built CRM for a niche the big platforms ignore — and wired Claude into the heart of it through a custom Model Context Protocol server, so the team can ask its business a question instead of hunting for the answer.
Selling vehicles to cities, counties, and law-enforcement agencies isn't ordinary retail. Deals move through cooperative purchasing contracts, every order carries procurement and compliance requirements, and a single sale can stretch across months of quotes, allocations, upfit work, and delivery milestones. The data that matters — receivables aging, inventory by lot, where each deal sits, which orders are stuck — lives in a dozen places and a hundred spreadsheets.
Generic CRMs weren't built for any of this. They model a sales funnel, not a fleet operation. The result was familiar to anyone in the niche: reps spending more time assembling the picture than acting on it, and leadership flying blind between manual reports. The market needed software that understood government fleet on its own terms — and there wasn't any.
True-Compass.AI built GovFleet.AI from the ground up for government fleet dealers: a system that speaks the language of customers, quotes, orders, deliveries, inventory, and receivables the way this business actually works. But the differentiator isn't the data model. It's that the whole operation became something you can simply ask.
We built a custom Model Context Protocol (MCP) server on Firebase Cloud Functions that exposes the CRM's live data to Claude through purpose-designed, access-scoped tools. Instead of clicking through reports, the team asks Claude — and Claude reaches into the real system to answer.
"What's in the over-90 receivables bucket?" "Which orders are waiting on delivery?" "Summarize this rep's pipeline." Plain-language questions, answered from live CRM data — no report-building required.
Receivables aging, inventory snapshots, pipeline summaries, quote metrics, call logs — each surfaced as a scoped, reliable tool so Claude pulls the right data instead of guessing at it.
Concept agents built on the same foundation — a pipeline nudge, a post-sale status check, a daily leadership briefing — turn the data layer into proactive help, not just answers on demand.
Built by a 29-year law-enforcement veteran turned fleet sales leader, GovFleet.AI reflects how cooperative contracts and procurement rules actually constrain a deal — not a generic funnel bolted onto a niche.
The design principle was simple: Claude should answer from the same live data the team works in, with no stale exports and no hand-built integrations to maintain. MCP made that clean.
Because the MCP layer sits directly on the CRM's own backend, every answer reflects the current state of the business. New capabilities ship by adding a tool, not by rebuilding an integration — which is why the same foundation now powers both on-demand answers and the proactive agents above.
"I spent 29 years in law enforcement and then in government fleet sales — I knew exactly where this business loses time. The goal with GovFleet.AI wasn't to add AI for its own sake. It was to make the whole operation answerable. Now you ask it a question, and Claude reaches into the real system and tells you. That's the difference between software that stores your work and software that does it with you."
