Daggy

Daggy lets runs statistical analyses on your own data, directly from an any AI interface. Since it's an MCP server, it connects directly to chat interfaces like Claude, ChatGPT or Copilot. Developers can also connect it to their agents to allow them to run statistical and data analyses.

What it does

Trains models. Bayesian regressions and hierarchical models in NumPyro, by MCMC or variational inference, with model comparison when you have competing stories about the data.
Answers what-if questions. Fit sales on price and ad spend and then ask what happens if you raise prices ten percent. Stress-test a portfolio with Monte Carlo simulation. Compare scenarios against a fitted model rather than a rule of thumb.
Works on your data. Storage is built in. You upload a CSV or write data from your assistant, and query it in SQL or plain English (DuckDB backend).
Keeps state. Datasets and trained models persist between sessions, so you can come back to a model next week and ask it something new.

What it looks like

Here's a full run in Claude: a what-if question about a portfolio under two rate-and-tariff scenarios, Daggy fitting the model and querying the data, and Claude writing up the comparison.

Claude prompt asking Daggy to stress-test a portfolio under two rate and tariff scenarios

Claude calling Daggy tools: listing datasets, reading table metadata, querying data, running a statistical model

Claude comparing the two scenarios with a second model run and writing up the findings

Claude showing the portfolio stress test results table computed by Daggy

1. Ask the question in plain English — Claude has Daggy connected as a tool.

Using it

Daggy is a remote MCP server, so there's nothing to install. In claude.ai or ChatGPT, add it as a custom connector at https://mcp.daggydata.com/mcp and log in when prompted. Step-by-step instructions on setup and data ingestion are on the setup page. Existing accounts log in at app.daggydata.com.

If you'd like to use it on a real problem and want a hand, get in touch.