Summary

Plain-English questions against ADX, with the answer sized to fit.

The result handling is what you notice: execute-query stops at 20 rows unless told otherwise, and the whole response is then squeezed into a character budget by dropping rows instead of truncating them, so what arrives is still valid, parseable data with a note that more exists. Decide two settings before auto-approving the query tool — write and management commands are allowed by default, and telemetry is always on unless you point it at your own OTLP collector.

What it is

A KQL client for Azure Data Explorer. Point it at a cluster URL and database and it lists tables and functions, shows their schemas, and runs queries — sizing every response so a wide result set does not swamp the conversation.

What you get

A connection to any ADX cluster URL and database, authenticated as the identity already logged in through the Azure CLI
Tables and functions in the current database listed, with column schemas, and a function's own code and parameters
KQL queries run with results capped at 20 rows by default and a limit parameter when you need more, plus a flag on the response when the results are partial
Results as JSON or as markdown tables, with KUSTO_MARKDOWN_MAX_CELL_LENGTH keeping one long text column from blowing the table wide
A character budget for the whole response — KUSTO_MAX_RESPONSE_LENGTH, default 12000 — met by returning fewer whole rows rather than cutting data mid-value, never dropping below KUSTO_MIN_RESPONSE_ROWS
A read-only mode: with KUSTO_ALLOW_WRITE_OPERATIONS=false only plain KQL and .show commands run, and .ingest, .drop, .create, .alter and the rest are rejected
A pre-filled GitHub issue link from report-issue, which needs no GitHub token and no live Kusto connection — so it still works when the connection is what broke

Requirements

The Azure CLI on the machine and a live az login session — queries run as that identity, so cluster access follows your own Azure permissions. It starts over stdio as npx -y kusto-mcp@latest. KUSTO_AUTH_METHOD selects azure-cli or azure-identity; KUSTO_QUERY_TIMEOUT (default 60000 ms) and KUSTO_CONNECTION_TIMEOUT (default 20000 ms) bound the calls; KUSTO_RESPONSE_FORMAT switches between json and markdown. Write and management commands are permitted unless KUSTO_ALLOW_WRITE_OPERATIONS=false. Anonymous usage telemetry is always on and carries no cluster, database, table or function names, no query text and no results; OTEL_EXPORTER_OTLP_ENDPOINT and OTEL_EXPORTER_OTLP_HEADERS route it to your own OTLP pipeline. MIT.

Setup effort

One command — npx -y kusto-mcp@latest

Kusto MCP Server