MDO Meets Claude: Ask Investment Questions in Plain English. We Handle the Code
We’re announcing something that changes the day-to-day for every MDO user: MDO is now available as an MCP server, connected directly to Claude. That means you can do real, institutional-grade factor research, screens, and backtests by asking, in plain English, and let MDO do what MDO does best.
No Python to write. No SQL to debug. No environment to configure. No “please regenerate that, the look-ahead bias snuck back in” loops with an LLM.
One prompt. A full factor research workflow.
In the demo above, the prompt is a single sentence: build a factor backtest on the Russell 3000, monthly rebalancing, 2010 through 2020, with five value factors and five growth factors. What comes back is the entire workflow: universe construction, point-in-time fundamentals, forward returns, information coefficients across horizons, hit rates, quintile spreads, t-stats, rolling IC charts, cumulative quintile wealth curves, and regime commentary on what worked, what didn’t, and why.
That isn’t a chatbot guessing at finance. MDO handles the data retrieval and the analysis. Claude handles the rest.
Why this matters: the part where AI struggles, MDO already solved.
Here’s the uncomfortable truth most teams have already discovered the hard way. General-purpose AI is impressive at writing code in the abstract. It is much less impressive when it has to write correct code against your live SQL or Snowflake data. That’s the part LLMs struggle with the most: the joins are wrong, the dates aren’t aligned, point-in-time logic gets dropped, look-ahead bias creeps back in, currencies aren’t translated, units aren’t normalized, the “as-of” date silently slides forward, and the answer looks plausible right up until the moment you stake a real decision on it.
That isn’t an AI problem you can prompt your way out of. It’s a data engineering problem.
MDO already solved that data engineering problem. Date alignment, look-ahead bias prevention, currency translation, unit normalization, trailing-twelve-month calculations, point-in-time logic, and vendor normalization across FactSet, S&P, LSEG, and OptionMetrics. All of it is curated into a structured, analysis-ready foundation that’s been hardened over years of real institutional use.
When Claude calls MDO through MCP, it isn’t writing fragile Python against raw vendor tables and hoping. It’s calling pre-built, audited workflows on an investment-ready data layer. The AI handles the conversation. MDO handles the math.
What changes for MDO end users
Until today, getting AI to do investment research on your data meant one of two things: either a quant on your team wrote and maintained the Python, or you trusted an LLM to write SQL against Snowflake on your behalf, and then you re-validated every answer because the cost of being wrong is too high.
With MDO + Claude over MCP, neither of those is the job anymore. The job becomes asking the right question.
Want to know which value factor held up across the 2013–2019 value drought? Ask. Want a sector-neutral, region-neutral global factor model rebalanced monthly with turnover and information ratio across the last decade? Ask. Want to compare FCF/EV against book-to-price, point-in-time, on the Russell 3000, with quintile spreads and Newey-West t-stats? Ask.
You don’t need to know that the underlying call has to apply preliminary-vs-restated date handling on Worldscope, or that returns need a +6/+12 month shift, or that overlapping forward returns require HAC standard errors. MDO already knows.
Correct, reproducible, auditable, by default
Three things matter for institutional asset management, and they don’t change just because an AI is in the loop. The output has to be correct: built on the data nuances that actually move backtest results. It has to be reproducible: point-in-time, so the model you ran today matches the model you deploy tomorrow. And it has to be auditable: every step traceable, every dataset identified, every assumption visible.
MDO’s Financial Intelligence Engine is built around all three. Claude doesn’t change that contract; it just makes the front door wider. Your portfolio managers, fundamental analysts, and research leads can now drive that engine in plain English, without the coding tax in between.
Already built. Ready today.
The MCP integration is live for MDO customers. If you’d like to put it on your own data, schedule a walkthrough and we’ll show you the same workflow on a universe of your choosing.
Factor research is one workflow. There are infinite more. When MDO is your data layer, AI in asset management isn’t a project. It’s already built.