The BigQuery half works on its own: read-only queries with `maximum_bytes_billed` as a cost ceiling, and INSERT, UPDATE, DELETE, CREATE, DROP and ALTER blocked outright. Everything memorable — caching that invalidates on table changes, column documentation, performance history — depends on a Supabase project you set up with the README's schema, so decide up front whether you want that second dependency. Row Level Security and per-user cache isolation are there for multi-tenant use.
A FastMCP server over BigQuery with an optional Supabase knowledge base behind it. The core is read-only SQL and schema access; the Supabase half adds result caching with dependency-based invalidation, business context on columns, query history and schema-evolution tracking.
- A read-only SQL query with caching and a cost ceiling — `execute_bigquery_sql`, taking `sql`, `maximum_bytes_billed`, `use_cache`, `user_id` and `force_refresh`
- Dataset and table discovery with the documentation attached — `get_datasets`, `get_tables`, `get_table_schema`
- Business context rather than just column types — `explain_table`, with usage statistics on request
- Query recommendations drawn from past usage — `get_query_suggestions`
- Performance analysis over query history, with recommendations — `analyze_query_performance`
- Schema changes tracked over time, with impact analysis — `get_schema_changes`
- Cache operations and a connectivity check — `manage_cache` with actions such as `stats` and `cleanup`, and `health_check`
- Three transports from one binary — `mcp-bigquery --transport http`, `--transport stdio`, `--transport sse`
- Server-Sent Events for live monitoring: system status, query execution and resource updates on separate streams
Python with `uv`, installed from a clone as `uv pip install -e ".[dev]"`. BigQuery configuration goes in `.env` as `PROJECT_ID` and `LOCATION`, with `KEY_FILE` pointing at a service-account key or Google Cloud SDK default credentials used instead. The enhanced features need a Supabase project — `SUPABASE_URL` plus `SUPABASE_SERVICE_KEY` or `SUPABASE_ANON_KEY` — and the tables the README's SQL creates: `query_cache`, `table_dependencies`, `schema_snapshots`, `column_documentation`, `query_history`, `query_templates`, `event_log` and `user_preferences`.