BigQuery MCP server optimized for quick navigation of larger projects and datasets.
Wellknown found it in public sources; nobody has proven control of it yet. Claiming takes one click if the repository is under your GitHub account, or a small file on your domain otherwise. Verified owners get the badge, 15-minute checks, status alerts, edits that outrank crawled data, and a ranking boost.
Agents can do it too: POST https://wellknown.network/api/v1/claims with {"agent":"bigquery-mcp-python","method":"well_known_file"} — machine-readable steps at claim.json, guide at /docs/claim.
Everything here was measured by our prober or read from a registry. Nothing is self-reported.
Attributed to the source that supplied each field. Treated as claims, not facts.
# 🗂️ BigQuery MCP Server Practical MCP server for navigating BigQuery datasets and tables by LLMs. Designed for larger projects with many datasets/tables, optimized to keep BigQuery spend low and LLM context small while staying fast and safe. - **Minimal by default**: list datasets and tables names; fetch details only when asked - **Navigate larger projects**: filter by name, request detailed metadata/schemas on demand - **Quick table insight**: optional schema, column descriptions and fill-rate to help an agent decide relevance fast - **Safe to run**: read-only query execution with guardrails (SELECT/WITH only, comment stripping) - **Cost-bounded by design**: metadata-first discovery, a `dry_run_query` cost estimator, and a hard per-query bytes-billed cap ## 🎯 Optimization priority Every tool and default in this server is designed around one explicit ordering. When choices conflict, earlier goals win: 1. **Minimize BigQuery cost first** — bytes scanned is what you pay for. Discovery (`list_dataset_ids`, `get_dataset_info`, `list_table_ids`, `get_table_info`) is **metadata-only and scans zero bytes**. `dry_run_query` estimates a query's bytes **without running it**. Every real query (`execute_sql`, `get_table_info` sampling) is capped by `maximum_bytes_billed` (default ~USD 0.50/query). Tool descriptions steer the model to filter on partition/cluster columns, select only needed columns, and use `LIMIT`. 2. **Then minimize LLM (token) cost** — list tools return **names only by default**, switching to full metadata only when `detailed=true`. Responses are compact, structured JSON so the agent spends few tokens deciding what's relevant before paying for a scan. 3. **Then minimize latency** — metadata calls run in threads and time out fast; list+search uses a bounded fetch multiplier. Latency is optimized **only where it doesn't increase BigQuery or token cost**. See [ARCHITECTURE.md](ARCHITECTURE.md#design-priorities-bigquery-cost--llm-cost--latency) for the m…
Mapped onto the structured taxonomy from declared text and observed tool names. Confidence shown for derived entries.
Every source is kept verbatim. Field changes are logged as events.