Summary

Reads the cache before it spends a query.

Two of the seven tools exist purely to avoid re-execution — fetch a result by id, or take the latest cached result for a saved query — and the execution call itself accepts a cache max-age. On a shared Redash instance backed by a production warehouse, that is the difference between an assistant being useful and being a load problem. Setting a default data source id removes the most common argument from the query call.

What it is

An MCP server over the Redash API. It covers the awkward part of Redash automation — submitting a query and polling until the job finishes — and the read paths around it, so an assistant can work with a company's existing data sources and saved queries.

What you get

SQL execution that blocks until the result is ready, with an optional data source id and a cache max-age — `execute_query_and_wait`
Data source discovery: list everything available, or read one in detail — `list_data_sources`, `get_data_source`
Saved query access: fetch one by id including its SQL text, or search saved queries by keyword with paging — `get_query`, `search_queries`
Results without re-running anything: fetch an existing result by its result id, or the latest cached result for a saved query — `get_query_result`, `get_saved_query_result`

Requirements

A Redash API key from Edit Profile in your Redash instance, plus the instance URL: `REDASH_API_KEY` and `REDASH_BASE_URL`. Optional: `DATA_SOURCE_ID` as the default data source, and `PORT` for the HTTP modes (3000 by default). Run it with `npx mcp-redash` after `npm link`, or as a Docker container. Three launch modes: stdio by default, Streamable HTTP with `--streamable-http` exposing POST, GET and DELETE on /mcp, and a deprecated SSE mode kept for legacy clients. The package is `mcp-redash` (1.0.0 in package.json).

Setup effort

One command plus a key — npx mcp-redash, then supply credentials