# ClickHouse MCP Server

> ClickHouse's official MCP server: run SQL, list databases and tables on a ClickHouse cluster (read-only by default), with optional embedded chDB queries.

- id: `clickhouse-mcp-server`
- category: MCP Servers (`mcp-servers`)
- url: https://github.com/ClickHouse/mcp-clickhouse
- repo: https://github.com/ClickHouse/mcp-clickhouse
- docs: https://clickhouse.com/docs/guides/use-cases/ai-ml/MCP
- status: active
- pricing: open-source
- license: Apache-2.0
- tags: mcp, clickhouse, database, sql, analytics
- related: Postgres MCP Pro (https://indexagentica.com/entries/postgres-mcp-pro/); MCP Toolbox for Databases (https://indexagentica.com/entries/mcp-toolbox-databases/); MotherDuck / DuckDB MCP Server (https://indexagentica.com/entries/motherduck-mcp-server/); MongoDB MCP Server (https://indexagentica.com/entries/mongodb-mcp-server/)
- sources: https://github.com/ClickHouse/mcp-clickhouse , https://pypi.org/project/mcp-clickhouse/ , https://clickhouse.com/docs/guides/use-cases/ai-ml/MCP
- added: 2026-10-19
- updated: 2026-10-19
- last verified: 2026-10-19
- featured in: Data analysis agent stack (https://indexagentica.com/stacks/data-analysis-agent/)
- maintainer: ClickHouse
- submitted by: agentica-curator
- page: https://indexagentica.com/entries/clickhouse-mcp-server/
- json: https://indexagentica.com/api/entries/clickhouse-mcp-server.json

Tools: `run_query` (with typed query parameters), `list_databases`, `list_tables` (paginated, with column metadata) and, with the optional `chdb` extra, `run_chdb_select_query` on the embedded chDB engine. Queries run read-only by default (`CLICKHOUSE_ALLOW_WRITE_ACCESS=false`); writes need an explicit opt-in and destructive statements such as `DROP` need a second flag, though the README notes the real security boundary is the ClickHouse user's grants. Defaults to stdio; HTTP/SSE transports require authentication by default and expose a `/health` endpoint. Install from PyPI as `mcp-clickhouse` (e.g. via `uvx`).