jDataMunch MCP
An MCP server that answers questions about CSV, Excel, Parquet and JSONL without pasting rows into context.
- 83 ★ on the repository
- Last commit
Extended access
Why this level
- The server reads the data files you point to and writes a local index to disk
- Query result rows enter the agent context, so they can contain personal data
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/jdatamunch-mcpDetects the agents on your machine, shows the access level and pins the version.
What it sends to the author: what it sends to the author: an anonymous savings counter — a random ID and aggregate token counts. no other data is sent, sending happens by default without consent.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server reads the data files you point to and writes a local index to disk
- Query result rows enter the agent context, so they can contain personal data
This is third-party code. Review the repository files before installing.
What it does
jDataMunch indexes a dataset once: it computes columns, types, cardinality, null rates and distributions and stores the profile on disk. After that the agent works with the data on demand rather than a copy in the prompt. `describe_dataset` and `describe_column` answer from the profile, `sample_rows`, `get_rows` and `get_distribution` fetch rows and distributions, `aggregate` groups and counts on the server, and correlations and joins across datasets are available. Indexing starts from chat: "using jdatamunch, index ./data/sales.csv" calls `index_local`. In the author benchmark on a 255 MB file with a million rows, pasting into the prompt costs about 111 million tokens, while an answer about the dataset shape with `describe_dataset` takes about 892. The server runs over stdio and connects through `uvx` with no install.
Who it is for. Analysts and developers who ask an agent to explore big tables and do not want to pay tokens for every row.
Good fit when
- A million-row CSV arrived and you must understand its columns, types and gaps without loading it into context
- You need counts by category and about 20 result rows instead of a dump of every row
- Two datasets must be joined and filtered down to specific rows
- Data sits in Excel or Parquet and the agent must read it the same way as CSV
Not a fit when
- You need to work with a database or SQL server: the server reads files and does not connect to a DBMS
- The server is needed in a commercial project without a separate license: there is no free permission for that case
- The table is tiny and fits in a prompt: profiling and indexing will not pay off
Example request
Using jdatamunch, index ./data/sales.csv, describe the dataset and tell me which columns have missing values, then count rows by region.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Python 3.10 or newer is required. Excel and Parquet need the extra `jdatamunch-mcp[excel,parquet]` packages. The license is dual: free for noncommercial use, while work at a commercial company or for clients needs a paid one-time license from jMunch LLC. The index and profiles are stored locally on disk. The token saving figures in the README come from one public Los Angeles police records dataset, and results differ on other data.
How to disable. Remove the server with `claude mcp remove jdatamunch` and, if you like, uninstall the package with `uv tool uninstall jdatamunch-mcp`. Indexes and profiles created by the server stay on disk and are deleted separately.
FAQ
How do I connect the server to Claude Code?
With `claude mcp add jdatamunch -- uvx jdatamunch-mcp`. There is no separate install: `uvx` fetches and runs the server on demand. After a restart `jdatamunch` appears in `/mcp`.
Which formats are supported?
CSV, Excel, Parquet and JSONL. Excel and Parquet need extra packages added through `--from "jdatamunch-mcp[excel,parquet]"`.
How do I index a file?
Ask the agent: "using jdatamunch, index ./data/sales.csv". It calls `index_local` and returns the dataset name, row and column counts and detected types.
Do I need a license?
Not for personal work. For commercial work, jMunch LLC sells it.
Does the agent read the whole table?
No. Questions about shape are answered from the profile, while filters, aggregates and joins run on the server and return only the result.
Discussion
No questions or reviews yet.
Related
MCP serversDaData MCP
dadata-mcp
DaData MCP server with 31 tools: address suggestions, company lookup by INN, phone and passport cleaning, reference lists
Official Data Commons MCP server: an agent finds statistical indicators and pulls observations for countries, regions and cities.
Jitsu MCP configures and debugs data pipelines: destinations, streams, functions, live events and syncs.
A DataHub MCP server: an agent searches tables and columns by metadata, traces lineage and finds analysts' SQL queries
Dremio MCP server: an agent browses tables, looks for patterns in the data and analyzes cluster workload.
A DaJet Metadata based MCP server that exposes 1C metadata structure and runs SELECT queries against the database
Comparison with similar tools
Cards that share the main search query, and how they differ.
README in short
The description starts with the problem: pasting a table into a prompt costs millions of tokens. The mechanism of indexing once and querying on the server, a benchmark on a million-row file, install through `uvx`, a quickstart, a feature list and the dual-license terms with prices follow.