jDataMunch profiles and queries tabular data on your machine. This document states exactly what runs locally, what leaves the machine, and how to turn each of those off.

Report security issues privately via GitHub Security Advisories rather than a public issue. Please include a reproduction, the version (pip show jdatamunch-mcp), and your platform. Security reports are triaged ahead of feature work.

Supported versions: the current 1.x release line receives security fixes.

jDataMunch never uploads your data. Files are parsed, profiled, and indexed locally. Queries — filters, aggregations, joins, SQL — execute locally against the local index. No dataset content, column name, file path, or row value is transmitted anywhere as part of normal operation.

Indexes are stored on disk under your home directory. Deleting that directory removes every trace of indexing.

Everything jDataMunch does beyond answering a tool call is listed here. All of it is visible, opt-out, and reversible.

• Anonymous savings counter. The server can contribute an anonymous delta to the public community meter at j.gravelle.us — a random install ID plus aggregate token counts. No data, no column names, no dataset names, no file paths, no queries, no PII. Counters only. Opt out with JDATAMUNCH_SHARE_SAVINGS=0.
• Startup import of the local embedding backend. When a native embedding provider is configured, the server imports that library at startup on the main thread, before serving. This is not polish: importing it later, on the worker thread a tool call runs in, deadlocks on the Windows loader lock and hangs the call forever. Nothing is downloaded and no network is touched — the import alone is what matters. Opt out with JDATAMUNCH_EAGER_EMBED_IMPORT=0.
• Local performance ledger — off by default. With JDATAMUNCH_PERF_TELEMETRY enabled, per-tool latencies are recorded to a local database that analyze_perf reads. It never leaves your machine. Off unless you turn it on; delete the file to erase it.
• Local index storage. Datasets and their profiles are stored under your home directory.

Beyond the user-invoked calls below, the base package makes no other network calls and leaves no persistent background processes. There is no scheduler and no background reporting.

Each of these happens only when you invoke it, never on import and never in the background.

• GitHub repository indexing. index_repo fetches data files from api.github.com using a token you supply. Private repositories require your own credential; jDataMunch stores nothing and proxies nothing.
• Embedding providers. When you configure one, embed_dataset and semantic search call that provider's API. Column names and sampled values are sent to the provider you chose — this is inherent to embedding them, and it is why no provider is configured by default.
• AI summaries. JDATAMUNCH_USE_AI_SUMMARIES calls your configured provider only when enabled.

⚠ Paid cloud providers are never called automatically. Setting a provider API key in your environment is not consent to bill it; jDataMunch requires an explicit opt-in before sending anything to a paid endpoint.

security.py enforces these on every call:

Response size is bounded by JDATAMUNCH_MAX_RESPONSE_TOKENS and JDATAMUNCH_MAX_ROWS, so a query against a very large dataset cannot flood a context window by accident.

The base install has none of these.

A default install with no configuration sends nothing but the anonymous counter, and that is one variable away from silent.