Creativity · MCP — server

MCP ClickHouse Server

The MCP ClickHouse Server exposes ClickHouse's lightning-fast columnar analytics engine to any MCP client. It lets Claude or another LLM list databases, inspect table schemas, and run analytical SQL queries against your ClickHouse cluster — ideal for ad-hoc exploration of logs, events, and time-series data at scale.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Python
Transports: stdio

Capabilities

Tools: list_databases, list_tables, describe_table, run_select_query
Resources: schema:// URIs for each table
Read-only mode recommended for safety

Install

pip install mcp-clickhouse

Configuration

{
  "mcpServers": {
    "clickhouse": {
      "command": "uvx",
      "args": ["mcp-clickhouse"],
      "env": {
        "CLICKHOUSE_HOST": "your-host.clickhouse.cloud",
        "CLICKHOUSE_USER": "default",
        "CLICKHOUSE_PASSWORD": "<password>",
        "CLICKHOUSE_SECURE": "true"
      }
    }
  }
}

Frequently asked questions

What is the MCP ClickHouse server?

A community-built MCP server that lets LLM clients run SQL queries and introspect schemas on a ClickHouse database. Useful for analytics copilots and log exploration.

Is writing supported?

The server emphasizes read-only SELECT queries by default. Treat it as a read-mostly analytics tool — point it at a replica or a least-privilege user for safety.

Does it work with ClickHouse Cloud?

Yes — set CLICKHOUSE_HOST to the cloud endpoint and CLICKHOUSE_SECURE=true. The same server binary works for self-hosted clusters.

Sources

ClickHouse MCP Server on GitHub — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20