GenPark Reservoir Sampling Stream Auditor

A local Python MCP server that provides reservoir sampling and quantile estimation over high-volume streams, aimed at LLM telemetry and token-stream auditing.

What you can do

Sample a stream with O(k) space using unbiased reservoir sampling, keeping a bounded representative subset of a large or unbounded stream rather than storing everything. Estimate quantiles over streamed values, such as token counts or latency figures, from the retained sample. Run as a self-contained Python standard-library implementation with no external pip dependencies, intended for in-memory, low-latency use in AI agent workflows.

When to use it

Useful when an assistant needs bounded-memory statistics over a large or continuous stream of LLM telemetry or token data, for example sampling records or estimating quantiles without keeping the full stream.

Requirements

Runs locally over stdio as a Python process (started with python mcp_server.py); the README states it uses only the Python standard library with no external pip dependencies. No authentication, environment variables or headers are declared.