Snowflake MCP Server

Snowflake · Data & Analytics Infrastructure

Official

Official Snowflake-managed MCP server for secure, governed data access. Query warehouses, read metadata, and build data agents with Cortex-backed governance.

data warehousesqlcortexgovernance

Status

Official

Auth

OAuth

Hosting

Remote

Community rating

Not rated yet

Open documentation →

Rate this server

How to connect Snowflake MCP Server

Step 1

Remote server — paste a URL

Snowflake MCP Server publishes a remote (Streamable HTTP) endpoint, so nothing is installed locally. Add it as a custom connector in Claude, ChatGPT or your IDE and paste the endpoint below.

Endpoint

https://<account>.snowflakecomputing.com/mcp

Client config snippet

{
  "mcpServers": {
    "snowflake-mcp": {
      "url": "https://<account>.snowflakecomputing.com/mcp"
    }
  }
}

Step 2

Authenticate with OAuth

Your client opens a browser sign-in with the vendor. Approve only the scopes you need, and use a read-only account first when the platform offers one.

Step 3

Verify, then widen access

Restart your client, confirm the server's tools appear, and run one read-only question first. Keep client-side approval prompts on for any tool that can write, trade or send. Confirm with your firm's compliance policy before connecting systems holding client data.

Prompts that use this server

Starter prompts from the Practice Stacks that rely on Data & Analytics Infrastructure.

Custom reporting query brief

A firm-level answer pulled from your warehouse.

Portfolio Reporting & PerformanceData & Analytics InfrastructureCustody & Clearing

Read-only: answer <question, e.g. AUM by custodian and household size> from my analytics warehouse, then spot-check the answer against the live custodial data for two or three sample accounts. Show the SELECT query you ran, the row counts, the result table, and a reconciliation line comparing warehouse totals to custodian totals with any variance called out. Use SELECT only — no inserts, updates, deletes or schema changes — and do not place or modify anything at the custodian. If a table or account you need is missing, say so instead of approximating.

Guardrail · Read-only credentials only. Never connect a warehouse with write access to a chat tool.

Warehouse schema map

A plain-English map of what data you actually have.

Data Infrastructure (Builder / Enterprise)Data & Analytics InfrastructureWealth Platforms

Read-only: map the warehouse schemas and tables relevant to <domain, e.g. client holdings> — table, purpose, grain, row count and how tables join — then trace each one back to the upstream wealth platform or custodial API field that feeds it. Output a lineage table of Warehouse table.column / Upstream source field / Grain / Last loaded, and flag tables with no identifiable upstream source or with fields the source no longer provides. Use metadata and SELECT queries only — no DDL, no writes, no schema changes.

Guardrail · Read-only credentials. Never connect a warehouse with DDL rights to a chat tool.

Data quality profile

Nulls, duplicates and outliers in the tables you rely on.

Data Infrastructure (Builder / Enterprise)Data & Analytics InfrastructureMarket & Financial Data

Read-only: profile <table or dataset> — null rates by column, duplicate keys, value ranges and obvious outliers — then validate a sample of its prices, identifiers or corporate-action fields against a market-data source for the same dates. Show the SELECT queries you ran, the results, and a comparison table of warehouse value vs. market-data value vs. variance. Do not write, delete or correct any data. Recommend fixes as a list for my engineer to review.

Guardrail · Read-only. Corrections belong in your pipeline, with review and version control.

Read the full connection guide →