Skip to content

v0.1.0 — Initial public release

Choose a tag to compare

View all tags
@hugonissar hugonissar released this 12 May 10:12
· 2 commits to main since this release
v0.1.0
03a4d19
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

First public release of the BigQuery Read-Only MCP Server — a self-hosted MCP server for Google BigQuery with hard table allowlists, per-query scan ceilings, and built-in rate limiting.

What's included

  • Read-only MCP server exposing two tools: get_table_schema and query_assessments
  • Multi-dataset and multi-table support via comma-separated env vars (positional pairing)
  • Hard scan ceiling enforced via BigQuery dry-run before every query
  • Result-row cap, SQL length cap, request body size cap
  • Token-bucket rate limiter with separate concurrency semaphores for queries vs. metadata
  • LRU + TTL dry-run cache, TTL schema cache
  • Admin endpoint for schema cache invalidation, behind a separate key
  • Dockerfile and Cloud Run deployment guide

Getting started

See the README for full setup. In short: enable the required GCP APIs, create a least-privilege service account, store API keys in Secret Manager, and deploy with gcloud run deploy.

Security

This is a security-sensitive tool. Report vulnerabilities privately via GitHub security advisories. See SECURITY.md for details.

Known limitations

  • Single region per deployment (Cloud Run constraint)
  • No prompt-injection scanning (put a model security gateway in front if needed)
  • No column-level masking (use BigQuery authorized views)
  • Schema cache is per-instance

Compatibility

Tested with Claude Desktop, Cursor, Claude Code, and the OpenAI Responses API. Should work with any client speaking streamable-HTTP MCP.

Full Changelog: https://github.com/hugonissar/BigQuery-Read-Only-MCP-Server/commits/v0.1.0

Assets 2
Loading