Skip to content
ClickStack MCP Server

ClickStack MCP Server

by ClickHouse
Released
Repository Docs
Featured Cloud & Infrastructure

Summary

ClickHouse's observability MCP server — investigate logs, traces and metrics with semantic tools rather than raw SQL, hosted or self-hosted.

Features

  • Semantic investigation tools for logs, traces and metrics instead of raw SQL
  • Event pattern clustering to surface recurring, new and volume-driving errors
  • Trend analysis across historical windows and correlation of attributes to anomalies
  • Trace inspection for slow requests and outliers
  • Persist findings as shareable dashboards and saved searches
  • Managed OAuth 2.0 endpoint on ClickHouse Cloud, bearer-token endpoint for OSS/BYOC

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it.
  3. Restart the agent — the "ClickStack MCP Server" tools will be available.
{
  "mcpServers": {
    "clickstack-mcp": {
      "command": "https://mcp.clickhouse.cloud/clickstack"
    }
  }
}

Description

ClickStack is ClickHouse's observability stack, built on HyperDX, and this is its official MCP server: it puts an incident investigation surface in front of an AI agent instead of a bare SQL prompt.

Why it is separate from the ClickHouse MCP server. ClickHouse's general-purpose MCP server exposes the database — an agent writes SQL and reads rows back. That works for BI-style questions but is a poor fit for debugging, where the useful moves are "cluster these errors", "show me what changed" and "find the slow traces", each of which becomes a long and error-prone query. ClickStack instead exposes higher-level semantic tools built for those moves. ClickHouse reports 25% fewer tool calls and 2.5x greater consistency against its internal benchmarks compared with exposing raw SQL alone.

What the tools do. Event pattern analysis clusters similar logs and traces to surface recurring errors, genuinely new ones, and whichever are driving volume. Trend analysis compares behaviour across historical windows. Trace inspection pulls slow requests and outliers, and correlation ties attributes to an anomaly. Findings do not evaporate at the end of the session: the agent can persist them as dashboards and saved searches that the rest of the team can open. It can also search and aggregate logs, traces and metrics directly, manage alerts, explore data sources and webhook destinations, and list teams and active contexts.

Deployment. On ClickHouse Cloud it is a managed endpoint at https://mcp.clickhouse.cloud/clickstack over streamable HTTP with OAuth 2.0, enabled from the Cloud console under Connect → Connect with MCP, which then hands you pre-built configuration strings for Claude, Cursor and Codex CLI. Open-source and BYOC deployments serve the same server from their own instance at <instance>/api/mcp using a Personal API Access Key as a bearer token, so self-hosted ClickStack users are not left out.

Requirements. A ClickHouse Cloud service with ClickStack enabled, or a self-hosted ClickStack deployment. The managed server was announced in June 2026 following the open-source release at Open House 2026.

Related MCP Servers

MCP: SigNoz MCP Server

by SigNoz

Featured

Query SigNoz metrics, traces, logs, alerts and dashboards in natural language — and create alerts, dashboards and saved views back, with deep links into the UI.

Cloud & Infrastructure
Browse all MCP servers →