
ClickHouse
Give your agent safe, read-only access to ClickHouse schemas, parameterized SELECTs, and query plans while you build or debug analytics features.
Overview
ClickHouse (mcp-clickhousex) is a Build-phase MCP server that provides read-only ClickHouse metadata, parameterized queries, and plan analysis for coding agents.
What is this MCP server?
- Read-only MCP access to ClickHouse (no destructive writes via server design)
- Metadata discovery for databases, tables, and columns
- Parameterized SELECT queries for safer agent-generated SQL
- Query plan analysis for performance debugging
- v0.7.0 stdio server via uvx; optional MCP_CLICKHOUSE_DSN or multi-profile config.json
- Server version 0.7.0
- Read-only MCP server
- Transport: stdio via uvx on PyPI
What problem does it solve?
Debugging ClickHouse from an agent session usually means copying schemas and SQL into chat without guardrails or explain plans.
Who is it for?
Builders with ClickHouse-backed products who want agent-assisted schema exploration and query tuning during development.
Skip if: Full write access, schema migrations, or teams without a ClickHouse instance to connect.
What do I get? / Deliverables
You register the read-only MCP server with a DSN or profiles so the agent can list metadata, run bounded parameterized SELECTs, and review plans in place.
- Agent-visible ClickHouse metadata catalogs
- Parameterized read-only query results within server limits
- Query plan analysis output for agent-assisted tuning
Recommended MCP Servers
Journey fit
ClickHouse wiring happens while shipping backends, event pipelines, or dashboards that depend on fast OLAP queries. The server is an MCP database bridge—metadata discovery and guarded queries belong on the integrations shelf rather than a one-off SQL client habit.
How it compares
Read-only database MCP integration, not a ClickHouse hosting skill or ETL pipeline template.
Common Questions / FAQ
Who is mcp-clickhousex for?
Indie developers and small teams using ClickHouse who want AI agents to explore metadata and run read-only queries safely during build and debug.
When should I use mcp-clickhousex?
Use it while integrating analytics storage, validating agent-generated SQL, or analyzing slow queries with plan tools before you ship.
How do I add mcp-clickhousex to my agent?
Run via uvx (mcp-clickhousex), set MCP_CLICKHOUSE_DSN or profile DSNs in config, and add the stdio MCP server block in Claude Code, Cursor, or compatible clients.