Introducing the Lumonic MCP: artifacts included
Kevin Hsu
Today the Lumonic MCP is generally available for every Lumonic customer.
Connect Claude or ChatGPT to your Lumonic workspace and ask your portfolio questions directly: which companies haven't submitted financials, or which borrowers are close to breaching a covenant this quarter. Every number in the answer clicks back to the source document it came from, down to the cell.
What it is
MCP (Model Context Protocol) is the open standard that lets AI assistants read from the systems you already use. The Lumonic MCP server gives your AI of choice read access to your workspace: periodic financials, covenant status, reporting status, and the KPIs you track. It isn't tied to one model. Claude and ChatGPT both work, along with any MCP-enabled client your firm has approved.
It’s been live since early this year, with early adopters across private credit and private equity using it every day.
What you can do with it today
Ask which companies haven't submitted their May financials and you get the list, each name with what's missing and how long it's been outstanding. Ask about covenants and you get each borrower's tested threshold next to its actual, so "which borrowers are offside right now" has a grounded answer. Ahead of quarterly review, generate a one-pager for every company you cover, or pull exposure across the portfolio by the dimensions you track.
And because so much of this work ends in Excel, the MCP meets you there. It writes live Lumonic formulas into your workbook, so the numbers refresh and stay traceable instead of going stale on paste. The reporting pack you assemble by hand every quarter becomes a workbook where you switch the company in a dropdown and refresh.
Lumonic is a PitchBook company, so if your firm also uses PitchBook there's a bonus: the two MCP servers run side by side in the same conversation, your portfolio financials from Lumonic next to live market data from PitchBook.
Why the numbers hold up
You don't need Lumonic to put AI to work. Your analysts are already in Claude or ChatGPT every day. What they're missing is portfolio data a model can be trusted with: point one at raw files and it guesses at field names, numbers come back nobody can verify, and analysts end up rebuilding them in a side spreadsheet. Lumonic's job is data that's structured and reviewed before a model ever reads it. That's what makes a firm AI-ready.
Every platform in this market will have an MCP soon, and some already do. An MCP inherits whatever data sits underneath it. Ours starts with extraction built for the documents portfolio companies actually send, reviewed by your team and tested against your covenants and checks. Every figure it returns carries a traceback link: click it and Lumonic opens the source document, down to the cell the number came from. It's also why the connector launches read-only. Auditability comes first, and write access will arrive carrying the same traceback guarantees.
What's next
Over the coming weeks we'll publish how customers actually use it, starting with a CFO who built his firm's reporting workbook in Claude in an afternoon. Each one ships with its artifact: the prompt and the template, ready to run.
Get started
Everything you need is at lumonic.com/mcp: the quickstart and prompt library, along with the security documentation your information security team will ask for. Your cluster admin enables the connector, then find Lumonic in the Claude connector directory or add it directly.
The first artifact in the series ships today too: [artifact one, deliverables tracker recommended, confirm with Kieran], with the video, the prompt, and the template. Run it as is, or tell us and we'll tailor it to your firm.
If you're not a customer and want to see it on your own portfolio, we'll run a proof of concept on a few of your companies.
