ClickHouse

ClickHouse MCP gives Claude direct access to your ClickHouse clusters for analytical queries. It's built by ClickHouse themselves, which means it understands cluster topology and can discover nodes automatically. You point it at a host, pass credentials, and Claude can run SQL queries, inspect schemas, and pull aggregated data without you writing a single line of glue code. I'd reach for this when I'm doing exploratory analysis on large datasets and want to ask questions in plain English rather than writing SQL by hand. The typical workflow: you describe what you want, Claude generates the query, runs it, and explains the results. It's particularly good for ad-hoc reporting where you'd otherwise be context-switching between a chat interface and a database client. The cluster discovery feature is useful if you're running distributed ClickHouse setups, though most single-node users won't notice it. Quirks: it's vendor-official, so it's unlikely to break with ClickHouse updates, but it's also narrowly focused on query execution. Don't expect it to manage migrations or handle schema changes beyond inspection. The environment variables are mandatory, which is sensible for security but means you can't easily switch between multiple ClickHouse instances without reconfiguring. No npm package yet, so installation is a manual clone-and-build process. Who shouldn't bother: if you're not already running ClickHouse, this won't convince you to start. It's a query interface, not a database management tool. If your ClickHouse usage is purely programmatic and you've already got a solid query layer, adding an LLM middleman might be overkill. But if you're doing regular analytical work and want to offload query construction to Claude, it's a clean, purpose-built tool that does exactly what it says.