Snowflake Connector Setup & RBAC

The Snowflake MCP connector exposes enterprise analytical warehousing capabilities to your agent workspace while maintaining data residency boundaries and compute cost controls.

Setup Variables

• Account Locator: Your unique Snowflake locator string (e.g. xy12345.us-east-1).
• Warehouse: The designated computing cluster for agent queries.
• Database & Schema: Target analytical workspaces.
• Credentials: Private key or username/password references hydrated server-side from the MakeYourMCP credential vault — never passed in plain text to the LLM.

Why Snowflake Needs RBAC in AI Agent Pipelines

Snowflake warehouses often contain an organisation's most sensitive analytical data — revenue forecasts, customer PII, and confidential product metrics. Giving an AI agent unrestricted Snowflake access creates two risks: (1) accidental exposure of sensitive columns through LLM context window leakage, and (2) runaway compute cost if an agent issues expensive full-table scans without constraints. MakeYourMCP enforces Snowflake role mapping, column masking, and query cost budgets through policy rules, keeping agent queries safe and cost-predictable.

Example RBAC Policy

version: "mcp/v1alpha1"
metadata:
  id: "policy_snowflake_analytics_agent"
  name: "Analytics Agent — Read-Only Warehouse Access"

rules:
  - id: "rule_snowflake_read"
    role: "data_analyst"
    resource: "snowflake_prod.analytics.revenue_summary"
    actions:
      - "read"
      - "query"
    constraints:
      row_limit: 5000
      compute_budget_credits: 2
    approval:
      required: false

  - id: "rule_snowflake_raw_tables"
    role: "data_analyst"
    resource: "snowflake_prod.raw.*"
    actions:
      - "read"
    constraints:
      masked_columns:
        - "email"
        - "phone_number"
      row_limit: 500
    approval:
      required: true
      approvers:
        - "data_governance_team"

Common Pitfalls

• No compute budget on warehouse queries: LLM agents can generate inefficient SQL (full scans, missing WHERE clauses) that consumes large Snowflake credit budgets. Always set a compute_budget_credits constraint on analytical agent roles.
• Granting access to raw schema instead of marts: Raw schemas often contain unsanitised PII. Scope agent access to data mart or reporting schemas where columns are already masked or aggregated.
• Sharing the same Snowflake role as human analysts: Agent roles should be separate from human analyst roles in Snowflake's native RBAC, with narrower privileges, so that a compromised agent token cannot escalate to human-equivalent access.

See Also

• PostgreSQL Connector Setup Guide — for transactional database access alongside Snowflake
• Salesforce Connector Setup Guide — CRM access for cross-functional agents
• Slack Connector Setup Guide — send query result summaries via Slack agents
• GitHub Connector Setup Guide — source code access for engineering agents
• Policy YAML Schema Reference — full policy declaration specification
• Full Documentation Index
• Understanding RBAC in MCP Gateways — the theory behind these policy rules
• Credential Vaulting for AI Workflows — how server-side hydration protects credentials
• MakeYourMCP Security & Trust Center — full security architecture overview

← Back to documentation index