We connected Claude to our internal tools. Here's what we learned before we ever pointed it at a client system.

AI & Operations · May 1, 2026 · 5 min read

There's a lot of noise right now about AI inside EPM. Most of it skips a step.

Before you connect AI to a client's Oracle or NetSuite environment, you should connect it to your own business first. Your ticketing system. Your project tracker. Your operating data. The places where you actually feel the friction every day.

That's what we just did. And it's already changing how we work.

What we built

Last week we stood up an MCP server that connects Claude — and any other AI assistant that speaks the same protocol — to the internal tools we use to run Elevate EPM. Support tickets, project status, client health, daily summaries. The stuff that used to live in five tabs and three dashboards.

Now I can ask Claude things like "What's on my plate today?" or "Give me the full picture on this client — open tickets, project status, recent activity" and get a real answer pulled directly from our systems. Not a guess. Not a summary of what I told it last week. The actual current state.

What is MCP, briefly

MCP — Model Context Protocol — is an open standard for letting AI models talk to tools. Anthropic released it last year. The reason it matters: instead of every AI assistant needing custom integrations into every system, MCP gives you one shape that any compliant AI can use.

You build the bridge once. Claude can use it. Other models can use it. Your future AI assistants can use it. That's a meaningful architectural choice for a small firm — we're not betting the farm on a single vendor.

Why we started with our own tools, not a client's

A few reasons, and I think they're worth naming.

You can't responsibly demo something you haven't lived with. EPM support and managed services involve sensitive client data, custom workflows, and real business decisions. The right place to learn the rough edges of AI-powered tooling is inside your own four walls — where the only person who pays the cost of a mistake is you.

The benefits show up immediately, on the operations side. Most of what consumes a Managing Director's day is coordination and keeping high-priority clients moving. Knowing what's open, what's blocked, what's coming up, who needs a reply. That kind of context-gathering is where AI helps a lot, and where the stakes of an early misstep are low.

It builds the muscle. By the time we connect this to a client's EPM environment, we'll have spent months learning how to design these tools well. Where the friction is. What questions actually get asked. What guardrails matter. That experience compounds.

What it actually changes day-to-day

The honest version: it's not magic. It's leverage.

Before, if I wanted a real status check on a client, I'd open the support tool, scan tickets, switch to the project tracker, check phase and status, hold all of that in my head, then write the email. Maybe 10 minutes of context-loading before I started typing.

Now I ask once, get a synthesized read, and start writing. The 10 minutes turn into 30 seconds. Multiply that across a day of client touchpoints, partner conversations, and follow-ups, and you get back real time.

It also means the quality of those touchpoints goes up — because I'm operating with current data, not what I remember from yesterday.

AI as a force multiplier for the humans doing the work — not a replacement for them.

What's next

The EPM connection is coming. We've designed the MCP architecture with that in mind from the beginning — same protocol, same patterns, just pointed at Oracle and NetSuite instances instead of internal systems.

Worth being clear about what this is not: we're not building an AI reporting tool, and we're not putting AI in the seat of an end user. The point is to make our Managed Services work better — faster diagnosis on tickets, sharper context for our consultants, more responsive support for our clients. AI as a force multiplier for the humans doing the work, not a replacement for them.

For the EPM side specifically, we recommend Caprus AI (caprus.ai). They've built an MCP server purpose-built for Oracle and NetSuite EPM — secure by design, fast, and properly architected for the way finance data actually moves. We've evaluated the space carefully; this is the one we're confident putting in front of clients.

If you're a finance leader watching the AI-in-EPM space and wondering where the substance is underneath the marketing — this is what substance looks like. Not a demo video. Real tools, used daily, by the people doing the work.

More to come...