Dataset Explorer MCP Server
Dataset Explorer MCP server lets an AI assistant inspect and analyze local tabular datasets through MCP tools.
Summarize datasets: features, missing-value counts, categorical/numerical columns, and data types.
Report dataset shape (rows and columns).
Compute mean and median for every numerical feature.
Inspect a specific column or target column in detail.
Find duplicate rows.
Report columns with missing values.
Find strongly correlated numerical feature pairs above a threshold (default 0.8).
Detect outliers.
Analyze local file paths; README supports CSV, TSV, Excel, JSON, and Parquet.
Integrates with ChatGPT through a remote Streamable HTTP MCP endpoint at /mcp, letting ChatGPT orchestrate the dataset-analysis tools directly instead of using Gemini. Supports ChatGPT file attachments via the openai/fileParams metadata: ingest_dataset downloads an attached CSV/TSV/Excel/JSON/Parquet file, runs it through the ingestion/hash/schema pipeline, and returns a dataset_id that ChatGPT reuses to call overview, shape, statistical summary, column/target inspection, missing values, duplicates, correlations, and outlier tools.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Dataset Explorer MCP Server Analyze the dataset and show me missing values and correlations."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Dataset Explorer — MCP server for local data analysis
Download on PyPI · Official MCP Registry · Glama directory
Ask your AI assistant questions about a dataset on your computer. Dataset Explorer does the calculations so your assistant can explain what is in the file, what needs attention, and which columns are related.
For example: "Explore my sales file. Are any values missing? Are there repeated rows? Which columns are related to revenue?"
Your original file stays unchanged. There is no website to host and no Gemini API key to set up.
Quick demo: check a small CSV
Save this as sample.csv on the computer running the MCP server:
score,double,group
1,2,A
2,4,A
3,6,B
3,6,B
,8,BPrompt: “Use Dataset Explorer to check /full/path/to/sample.csv for missing
values and duplicate rows.” Replace the path with the file's actual absolute path.
MCP calls: analyze_missing_values({"path": "/full/path/to/sample.csv"})
and duplicate_finder({"path": "/full/path/to/sample.csv"}).
Reproduced result: 5 rows; score is missing in one row (row index 4,
20% of rows), and one row is a duplicate (20%). I reproduced these values by
calling both tools through the local MCP stdio client. The server returns JSON;
the assistant can turn it into a short explanation.
To try it yourself, install the package and connect a client. Earlier experiments show the Gemini-connected MCP client and the Streamable HTTP endpoint prototype. The installable version below runs locally over stdio.
Related MCP server: Vibe Preprocessing and Analysis MCP Server
What can it help with?
Summarize your data, including row counts, columns, averages, and medians.
Find missing values, repeated rows, and unusual numbers.
Look closely at a column you care about.
Compare columns and explore their relationships.
Supported files: CSV, TSV, Excel (.xlsx and .xls), JSON, and Parquet.
For Excel, the first worksheet is used. JSON files should contain table-like data.
Get started
1. Install it
You need Python 3.10 or newer. Run this in a terminal:
python -m pip install dataset-explorer-mcpThis downloads Dataset Explorer and the libraries it needs. You can also find it on PyPI.
2. Connect your AI assistant
Use Claude Desktop, Cursor, or VS Code with Copilot. These apps can connect to local tools through MCP. Choose your app below for the settings to add.
For Claude Desktop, open Settings > Developer > Edit Config.
For Cursor, use its MCP settings or the .cursor/mcp.json file in your project.
Add this entry to your configuration, keeping any servers you already have:
{
"mcpServers": {
"dataset-explorer": {
"command": "python",
"args": ["-m", "mcp_server"]
}
}
}Create or open .vscode/mcp.json in your project and add this server:
{
"servers": {
"dataset-explorer": {
"type": "stdio",
"command": "python",
"args": ["-m", "mcp_server"]
}
}
}Restart your app or reload its MCP settings. It will start Dataset Explorer for you. If it cannot find Python, see the quick fixes below.
3. Ask about a file
Give your assistant the full path to the file and ask your question:
Use Dataset Explorer to summarize
C:/Users/YourName/Downloads/sales.csv. Check for missing values and repeated rows, then explain the results.
On macOS or Linux, use a path such as /Users/yourname/data/sales.csv or
/home/yourname/data/sales.csv.
The file must be on the computer where the server runs. Your assistant may send the calculated results to its AI provider according to that app's settings.
How does it find relationships?
It calculates established statistics directly from your dataset:
Pearson correlation: how two numeric columns move together.
Eta squared: how numeric values differ between groups.
Cramer's V: how two category columns are related.
It chooses the method to suit the columns. These results help you spot patterns; they do not prove that one thing causes another. Results depend on the data you provide.
Quick fixes
Python or the package cannot be found: run
python -c "import sys; print(sys.executable)"in the terminal where you installed it. Use the printed path in place ofpythonin your app's configuration. On Windows, use forward slashes in that path.File not found: give the full file path and check that your assistant has permission to read it.
No tools appear: reload the MCP settings and check your app's server logs.
The terminal seems idle: that is normal. This server waits for your assistant to connect; it does not open a chat window of its own.
Large files need enough memory because the server loads the dataset for each request.
The server uses local stdio. Start it with dataset-explorer-mcp or
python -m mcp_server. Logs go to stderr; stdout carries MCP messages.
Its ten tools are get_dataset_overview, dataset_shape,
dataset_statistical_summary, inspect_Column, analyze_target,
duplicate_finder, analyze_missing_values, find_correlations,
detect_outliers, and screen_target_relationships. Every tool accepts a local
file path. A guide is available at dataset://guide, along with an
explore_dataset prompt.
After cloning this repository:
python -m pip install -e ".[dev]"
python -m pytest -q
python mcp_server.pyBuild a package with python -m build.
More of my work
Veritas: FastAPI verification and human-review workflow for AI-generated claims.
FlyRank ML internship: DuckDB page-pair features and a weighted-overlap baseline.
Haq team write-up: a voice-first legal guidance hackathon prototype.
MIT licensed. See LICENSE.
Available Tools
9 toolsanalyze_missing_valuesC
Reports columns with missing values in the dataset.
Inputs: path (required).
analyze_targetC
Gives a detailed summary of the target column.
Inputs: path (required), target_name (required). Returns: object
dataset_shapeB
Returns the number of rows and columns in a CSV dataset.
Inputs: path (required). Returns: object
dataset_statistical_summaryA
Returns the mean and median of every numerical feature.
Inputs: path (required). Returns: object
detect_outliersC
Detects Outliers in the dataset.
Inputs: path (required).
duplicate_finderC
Reports duplicate rows in the dataset.
Inputs: path (required). Returns: object
find_correlationsA
Finds strongly correlated numerical feature pairs above a specified absolute threshold.
Inputs: path (required), threshold. Returns: object
get_dataset_overviewB
Returns dataset features, missing-value counts, categorical columns, numerical columns, and data types.
Inputs: path (required). Returns: object
inspect_ColumnC
Gives a detailed summary of a specific column/feature.
Inputs: path (required), col_name (required). Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 9 tools
Some overlap exists between get_dataset_overview and analyze_missing_values, as both report missing-value counts. Additionally, inspect_Column and analyze_target are similar in nature, though they target different columns. Most other tools are clearly distinct.
Naming is inconsistent: some tools use verb_noun format (get_dataset_overview, analyze_target, find_correlations) while others use noun phrases (dataset_shape, duplicate_finder). The tool inspect_Column mixes snake_case with camelCase.
Nine tools is well-scoped for a dataset exploration server, covering essential analysis tasks without redundancy.
The tool set covers a broad range of dataset exploration needs including shape, stats, missing values, duplicates, correlations, and outliers. A data preview or column value counts would be a minor addition, but no critical gaps are apparent.
Maintenance
Related MCP Connectors
Ask questions about a CSV in plain English and get answers, charts and cleaned data back.
Dedupe, compare, merge, split and profile CSV or TSV files that are too big to paste into a chat.
Open, inspect, filter, edit and convert xlsx and csv files from your AI chat. Processing is local.
Open, inspect, filter, edit and convert xlsx and csv files from your AI chat. Processing is local.
Related MCP Servers
- AlicenseBqualityDmaintenanceEnables autonomous data exploration on .csv-based datasets, providing intelligent insights with minimal effort.2544MIT
- FlicenseNot gradedqualityDmaintenanceEnables users to preprocess, analyze, and visualize CSV data through comprehensive tools for data manipulation, statistical analysis, and graph generation.3-
- FlicenseAqualityDmaintenanceEnables comprehensive analysis of CSV files and SQLite databases through tools for statistics, correlations, anomaly detection, pivot tables, time series analysis, visualization, and automated insights discovery.16-
- FlicenseAqualityDmaintenanceEnables exploratory data analysis and machine learning on CSV datasets with tools for profiling, missing values, correlation, plotting, model training, and prediction.8-