Data Profiler MCP server

<!-- mcp-name: io.github.Ridadata/mcp-data-profiler -->

mcp-data-profiler

Let an AI agent understand a dataset without reading it.

An MCP server that turns a CSV, Parquet, JSON, or Excel file into a compact structured profile — types, ranges, missing values, and likely data-quality problems — instead of raw rows.

Overview

To let an AI agent reason about a data file, you normally paste rows into the conversation. That is expensive, truncates on anything large, and still leaves the model guessing at column types and null rates.

This server answers the question directly. One tool call returns a structured summary that is orders of magnitude smaller than the data and says more about it:

The 645 MB file cannot go into a context window at any price. It is fully characterised here in under two seconds.

Demo

python demo.py generates a deliberately messy dataset, profiles it, and reports what came back:

The same profiler seen from an MCP client — one question, one profile_dataset call, and the file is characterised without a single row entering the conversation:

Three real problems surfaced before any analysis began: a column that never varies, one that is entirely empty, and a date column that sorts as text — so "2024-10-01" < "2024-9-01" — silently corrupting any time-based result.

<!-- Both recordings are VHS tapes — docs/demo.tape and docs/mcp-demo.tape — and regenerate with two docker run lines. The conversation in the second one is scripted, but every figure in it comes from a real profile_dataset call; see docs/vhs/README.md. -->

Features

stored as text.

• Six data-quality flags — constant, all-null, probable ID, mixed types, and numbers or dates

quartiles for numerics, ranges for dates, frequent values for categories.

• Full column statistics — dtype, null count and percentage, distinct count, sample values,

what it truncated.

• Bounded output — the response stays small no matter how wide the input, and always reports

included.

• Honest sampling — large files are sampled, but never silently; the true row count is always

.ndjson .xlsx .xlsm .xls, plus .gz variants of the text formats.

• Five formats, eleven extensions — .csv .tsv .txt .parquet .pq .json .jsonl

• Path confinement — optional --root restricts profiling to a single directory.
• Zero configuration — no database, no index, no warm-up. Point it at a file.

Shortened. The full README is on GitHub.