mcp-bigquery-server MCP — Cas d'usage, Installer & Démo en direct

Pourquoi l'utiliser

Fonctionnalités clés

Read-only SQL — no INSERT/UPDATE/DELETE/DDL
Configurable query scan limit (default 1 GB) to avoid runaway costs
Field-level restrictions — list sensitive columns; agent can aggregate but not read rows
Schema exploration with table/view differentiation
Service account JSON or gcloud CLI auth

Démo en direct

Aperçu en pratique

bigquery-server.replay ▶ prêt

0/0

Installer

Choisissez votre client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "bigquery-server": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-bigquery-server"
      ],
      "_inferred": true
    }
  }
}

Ouvrez Claude Desktop → Settings → Developer → Edit Config. Redémarrez après avoir enregistré.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "bigquery-server": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-bigquery-server"
      ],
      "_inferred": true
    }
  }
}

Cursor utilise le même schéma mcpServers que Claude Desktop. La config projet l'emporte sur la globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "bigquery-server": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-bigquery-server"
      ],
      "_inferred": true
    }
  }
}

Cliquez sur l'icône MCP Servers dans la barre latérale Cline, puis "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "bigquery-server": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-bigquery-server"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "bigquery-server",
      "command": "npx",
      "args": [
        "-y",
        "mcp-bigquery-server"
      ]
    }
  ]
}

Continue utilise un tableau d'objets serveur plutôt qu'une map.

~/.config/zed/settings.json

{
  "context_servers": {
    "bigquery-server": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp-bigquery-server"
        ]
      }
    }
  }
}

Ajoutez dans context_servers. Zed recharge à chaud à la sauvegarde.

claude mcp add bigquery-server -- npx -y mcp-bigquery-server

Une seule ligne. Vérifiez avec claude mcp list. Supprimez avec claude mcp remove.

Cas d'usage

Usages concrets : mcp-bigquery-server

Answer product/growth questions from BigQuery without writing SQL

👤 PMs, growth analysts with BQ-backed warehouse ⏱ ~15 min intermediate

Quand l'utiliser : You have a question whose answer lives in events tables in BQ.

Prérequis

GCP service account with BQ Data Viewer + Job User — IAM > Create service account; download JSON key

Déroulement

Discover tables

List tables in dataset analytics. Describe events and users.✓ Copié

→ Schemas
Ask the question

How many users who signed up in March 2026 triggered the 'aha_moment' event within 7 days?✓ Copié

→ Numeric answer with SQL shown
Caveat

Any caveats? Timezone, deletion, test users?✓ Copié

→ Honest caveats

Résultat : Answers in minutes instead of data-team tickets.

Pièges

Running SELECT * on a huge fact table blows the scan limit — Always filter by partition column (often _PARTITIONDATE)

Let a less-trusted analyst explore data without reading PII rows

👤 Data platform teams ⏱ ~30 min advanced

Quand l'utiliser : You want to open BQ access to more people via chat without each of them being able to read customer emails.

Déroulement

Configure restricted fields

Add config.json entry restricting fields users.email, users.phone, users.ssn. Agent can only aggregate these, not SELECT them raw.✓ Copié

→ Config in place
Test

Run SELECT email FROM users LIMIT 10. Verify it's blocked. Then run SELECT domain, COUNT(*) FROM users GROUP BY domain — verify it works.✓ Copié

→ Block on raw read; allow on aggregate

Résultat : Safer self-service analytics for the LLM era.

Pièges

Regex-based field detection can miss complex aliased SQL — Defense in depth — also use BQ column-level security / authorized views

Combiner avec : gateway

Auto-compile a daily metrics digest from BQ

👤 PMs, founders ⏱ ~30 min intermediate

Quand l'utiliser : You want KPIs in Slack every morning without a BI tool.

Déroulement

Define the metrics

Define queries for: DAU, signups, revenue, top-3 errors. Each with yesterday / 7-day avg.✓ Copié

→ SQL per metric
Run and format

Run all and format as a Slack-ready digest. Include week-over-week deltas.✓ Copié

→ Slack-ready message

Résultat : Daily metrics without managed BI cost.

Combiner avec : notion

Combinaisons

Associez-le à d'autres MCPs pour un effet X10

bigquery-server + notion

Weekly KPI doc

Run my weekly KPI queries and create a Notion page in 'Metrics Weekly' with results + commentary.✓ Copié

bigquery-server + gateway

PII-safe access via mcp-gateway + Presidio

Put BigQuery MCP behind mcp-gateway with Presidio; verify customer emails get redacted in results.✓ Copié

Outils

Ce que ce MCP expose

Outil	Entrées	Quand appeler	Coût
list_datasets		First step to orient	free
list_tables	dataset	Navigate a dataset	free
describe_table	dataset, table	Before querying	free
query	sql: str, max_bytes?: int	Main read tool; limited to 1GB scan by default	BQ on-demand: $6.25 per TB scanned

Coût et limites

Coût d'exécution

Quota d'API: BigQuery job quotas (generous)
Tokens par appel: Query results can be huge — always LIMIT or aggregate
Monétaire: Pay GCP by bytes scanned ($6.25/TB on-demand). Configure scan limit in MCP to cap.
Astuce: Filter by partition. A full-table scan on a busy fact table = real money. The MCP's byte limit is your safety net.

Sécurité

Permissions, secrets, portée

Portées minimales : bigquery.dataViewer + bigquery.jobUser on specific datasets only

Stockage des identifiants : Service account JSON in a mounted path; never commit

Sortie de données : Query results go to your LLM provider

Ne jamais accorder : dataOwner / dataEditor to the MCP's service account

Field restrictions are cooperative guardrails for agents — not a replacement for BigQuery IAM/column-level security.
Developer preview status noted by upstream — verify behavior in your setup before prod use.

Dépannage

Erreurs courantes et correctifs

PERMISSION_DENIED on dataset

SA lacks BQ Data Viewer. gcloud projects add-iam-policy-binding ....

Vérifier : gcloud bigquery datasets list

Query exceeds configured byte limit

Add partition filter or column projection; or raise limit if legitimately needed.

Restricted field still appearing in results

Regex match may miss aliased columns — use BQ authorized views for hard isolation.

Alternatives

mcp-bigquery-server vs autres

Alternative	Quand l'utiliser	Compromis
Looker / Metabase	You want a BI tool, not chat	Better dashboards; less conversational
postgres MCP via Cloud SQL	Your analytical data is in Postgres instead	Different engine; Postgres doesn't scale like BQ for big aggregates

Plus

Ressources

📖 Lire le README officiel sur GitHub

🐙 Voir les issues ouvertes

🔍 Parcourir les 400+ serveurs MCP et Skills