list_databases, list_tables and get_table_info let an agent map the cluster before writing a query, which is where the failed attempts otherwise go. Hydrolix's own advice is a service account token with a read-only role — the grants on that token are the real boundary, not the tool list.
Hydrolix's own server for its log-analytics clusters. Four tools over one cluster connection: the databases it holds, the tables inside them, one table's schema, and a SQL query of your own.
- Every database on the cluster — list_databases
- The tables in a database — list_tables
- One table's metadata, schema included — get_table_info
- A SQL query executed against the cluster — run_select_query
- A /health endpoint under HTTP or SSE transport, returning the query head's ClickHouse version, or 503 when the cluster cannot be reached
HYDROLIX_URL pointing at your cluster, plus credentials: either HYDROLIX_TOKEN, a service account token, or HYDROLIX_USER and HYDROLIX_PASSWORD together. Ships on PyPI as mcp-hydrolix 0.3.6 and runs under Python 3.13, usually launched through uvx. Under HTTP or SSE the token can instead travel per request, as a bearer header or a ?token= parameter.
One command plus a key — pip install mcp-hydrolix, then supply credentials