Starting today, you can point Claude, ChatGPT, or agents you build yourself directly at the engine that runs your fund.

Here's the pattern I see at almost every firm we talk to. AI is everywhere, and it's making everyone slightly faster: faster drafts, faster lookups, faster summaries. And a year in, the work on everyone's desk is the same work. The reason is simple: an agent can only take over work you trust it to run on its own, and how your fund works (the fee bases, the waterfall mechanics, what that side letter actually means) lives in your people's heads and their spreadsheets, not in a system an agent can operate.

We spent five years building that system. Today we're opening it to your agents.

What we shipped

MCP, the Model Context Protocol, is the open standard AI agents use to connect to the systems where work actually happens. Today we're releasing ours. Connect once, and the AI tools your firm already uses can work with Maybern directly. No exports, no rekeying, no integration project.

Your agent reaches the engine itself: your funds' actual waterfall, fee, and performance logic, configured to your documents, doing the math the same way every time. And every answer comes back with its lineage. You get past "this was the number" to "this is how we got the number."

What you can do this week

• An LP meeting tomorrow. Ask your agent for a performance dashboard built from live numbers, and because it reaches Maybern alongside your CRM and everything else it's connected to, the fund data and the relationship context land in one deliverable.
• A prospective LP's track-record request, in their template. Hand the agent the template; it fills the performance tabs from your current numbers and flags anything that needs partner judgment.
• A fund admin package to tie out. Hand the agent the admin's support file. It reconciles line by line against Maybern and hands you the breaks, each one with the reason behind it.
• An auditor asking for support. The agent builds a working Excel file, formulas included, that ties out to the published numbers in Maybern, ready for your review before it goes back out.
• A follow-up you'd normally punt on. "Build the bridge from gross to net IRR for Fund II," answered in minutes, with every step traceable to source.

The engine underneath is the whole story

Every fund platform will ship an MCP eventually. What matters is what the agent finds on the other end of the connection.

I've spent a lot of late nights over the past few months working alongside agents on real fund questions, and the pattern is consistent. The first question is never the hard part. Any AI hooked up to any data can answer "what's our net IRR?" The divide shows up on the follow-up: why did carry move this quarter? What's driving the difference between our gross and net? How did you get that?

Follow-up questions can't be looked up. Nobody stored the answer, because nobody knew you'd ask. The answer has to be worked out, with your firm's actual calculation logic, on your firm's actual numbers.

That's why we spent years building MXL, a custom formula language that stores the work behind every number it produces. The answer isn't in the LPA; LPAs are often written ambiguously on purpose. The answer is in your firm's interpretation of the LPA, and at most firms that interpretation lives in people's heads. In Maybern it's written down in MXL: every fee basis, waterfall tier, and side-letter term as an expression the engine runs, reviewed once, versioned, live from then on. When your agent answers the follow-up, it isn't improvising an interpretation. It's running yours, and it can show you every step.

That's the line between an AI that can look up your numbers and one that can work them out. We built Maybern for GPs who ask the follow-up question.

Taking work off your desk

Where a firm enables it, agents go past reading: they can author new calculations and build reports.

An investor asks for 1-, 3-, and 5-year IRRs from liquidation. That's now a message to Claude: the agent writes the calculations in MXL, produces the support, and your team reviews and activates them.

Every write lands as a draft. Nothing becomes final until a person completes it in Maybern. And we're shipping more write capabilities steadily. Staging draft capital activity is next, each one moving another category of work from "the agent made me faster" to "the agent took it off my desk."

Where this goes

Maybern is the operating system for agentic fund management: the engine that stores your numbers and the logic behind them, configured, versioned, and queryable, and the agents that run on top of it. Today's release opens the engine to any agent you choose. In the coming months we're shipping our own agents off the shelf: reconciliation that runs when your fund admin's package lands, distribution drafting, audit support assembled before the auditor asks, each one delivering finished work to your team for approval. And our Forward Deployed Engineers are already building agents shaped to how individual firms run, on the same MCP you can connect to today.

Every firm's AI has been stuck at "slightly faster" for the same reason: nothing it could reach knew how the fund works.

Starting today, something does.

The Maybern MCP is available to customers now. Talk to your account team, or start at maybern.com/solutions/mcp.