shubham303/meelu-analytics-mcp

Local statistical analysis of CSV datasets with 45 tools — profiling, association testing, feature engineering, clustering, supervised ML with held-out scoring, time-series decomposition and forecasting, causal inference, and customer analytics (RFM, retention cohorts, market basket). Method selection is deterministic: it checks normality, variance, and expected cell counts, routes to the appropriate test (Pearson/Spearman, Welch/Mann-Whitney, ANOVA/Kruskal-Wallis, chi-square/Fisher), and records which it chose and why; every result carries a confidence level and declines when the data cannot support the question. DuckDB-backed sessions persist across restarts and data never leaves the machine. Streamable HTTP at `http://127.0.0.1:8321/mcp` (or stdio with `--stdio`); no auth, no API key. Clone the repo, then run `uv run --project . meelu-analytics-mcp`. Python 3.10+, macOS/Windows/Linux, MIT.

Project

Browse this MCP server in the TensorBlock MCP Index.