Introducing meelu-analytics: deterministic data analysis over MCP

Quick answer

meelu-analytics-mcp is a free, open source MCP server that turns Claude, Cursor or any MCP-capable agent into a careful data analyst for CSV files. It loads your data into a local DuckDB database, picks statistical tests deterministically, and returns every result with a confidence level and caveats — refusing to answer when the data can't support the question. Get it at github.com/shubham303/meelu-analytics-mcp.

meelu-analytics-mcp is the first open source piece of Meelu, and it's out today. Here's the problem it solves and how it works.

The problem

Ask an LLM to analyze a spreadsheet and it will happily do so — by reading the rows into its context window and eyeballing them. That fails in two ways. First, context windows are finite: a few thousand rows and the model is summarizing samples, not analyzing data. Second, the statistics are vibes. The model picks a number that looks plausible, and there is no way to tell a real finding from a hallucinated one.

What meelu-analytics does differently

The server ingests your files into DuckDB and runs the analysis itself. The model never sees your rows — it sees results.

• Sessions over files. You point it at CSV paths, it creates a session, detects foreign-key relationships between tables, and everything after that runs as SQL and statistics inside DuckDB.
• Deterministic method selection. When you ask whether two columns are associated, the server picks the right test based on the column types and data shape, not on what the model feels like. Same data, same question, same method, same answer.
• A trust block on every result. Every answer carries a confidence level (high / moderate / low / none) and explicit caveats. Small sample? Violated assumption? It says so.
• It refuses. When the data cannot support the question, the result comes back declined with a reason, instead of a made-up number. We think this is the single most important feature.

Beyond descriptive statistics it does clustering, outlier detection, changepoint detection, forecasting, retention cohorts, RFM segmentation, market-basket analysis, and trains small classifiers and regressors — each with the same structured, trust-annotated output.

Why we're releasing it

Meelu is a desktop marketing desk for founders, and analytics is one of its organs. But nothing about this server is marketing-specific — it's useful for any tabular data you'd rather not paste into a chat window. So it ships standalone, today, while the rest of Meelu is still in development.

Install it, point your agent at a CSV, and ask a question you already know the answer to. Then ask one you don't.