jgravelle/jdatamunch-mcp

claude mcp add jdatamunch -- uvx jdatamunch-mcp
claude mcp add jdatamunch -- uvx --from "jdatamunch-mcp[excel,parquet]" jdatamunch-mcp
uv tool install jdatamunch-mcp
pipx install jdatamunch-mcp

Usage

• •No separate indexing command: ask the agent to index a file and it calls index_local, returning the dataset name, row/column counts, and detected types.
• •Orient with describe_dataset / describe_column, then query without loading rows via get_rows, aggregate, or run_sql.
• •Work across datasets with suggest_joins, suggest_keys, and join_datasets; index CSVs/Excel/Parquet/JSONL straight from GitHub with index_repo.
• •Requires Python 3.10+ and any MCP-compatible client; Excel and Parquet need the matching optional extras.

Key features

• ✓39 tools covering orientation, server-side query, data quality (get_dataset_health, get_data_hotspots), schema-drift checks, and semantic search.
• ✓Aggregations and filters execute against stored data rather than being simulated in the model.
• ✓Sampled statistics report error bounds (roughly 2% standard error) instead of presenting estimates as exact.
• ✓Local-first with opt-out anonymous savings counter; no data, column names, file paths, or PII uploaded.
• ✓Implements the jMRI retrieval spec with token accounting shared across the jMunch suite.

Best for

Analysts and coding agents that need to explore large tabular datasets (CSV/Excel/Parquet/JSONL) without dumping rows into the context window.

Caveats

• ⚠Free for personal use; commercial use requires a paid one-time license (from $39) sold by jMunch LLC.
• ⚠Token savings scale with file size — negligible on small spreadsheets.
• ⚠Excel and Parquet support require optional extras that pull extra dependencies.
• ⚠Does not read code or prose; those belong to the sibling jcodemunch-mcp and jdocmunch-mcp servers.
• ⚠Default describe_column is not labeled offloadable when index freshness is unknown.