---
title: "Snowflake MCP: named agents, logged calls | Rig"
description: "Snowflake MCP with its own role, token and call limit for each agent. Govern Snowflake access per agent and reduce credit burn with Rig, with every query logged."
canonical: "https://rig.so/mcp/snowflake"
format: markdown
---

MCP

# Every Cortex call needs a name and a log

Snowflake MCP lets Claude, Cursor and your own agents query Snowflake through the Model Context Protocol. Rig puts Gateway in front of it, so you can govern Snowflake access per agent and reduce credit burn: each agent gets a role, a token, a daily call limit and the rows its role allows, and every call it makes is logged under its name.

By [Rig](https://rig.so/about) · Last updated 20 November 2026

[Book a call with an architect](https://rig.so/book-demo)

## At a glance

- Snowflake runs a managed MCP server, created with CREATE MCP SERVER, that exposes Cortex Agents, Cortex Analyst, Cortex Search and SQL execution as tools.
- Access to that server is granted per tool, and sessions use the caller’s default role, so agents need roles of their own.
- Rig’s Gateway gives each agent its own role, token, daily call cap and a named audit trail, with certified metrics from Rig Map looked up before any SQL runs.
- Context lookups come from Rig Map, and the daily cap stops an agent stuck in a loop, which cuts the warehouse queries an agent runs.

## What is Snowflake MCP?

MCP is an open protocol that lets an AI tool discover and call tools on a server. Snowflake’s managed MCP server is a Snowflake object you create with CREATE MCP SERVER and a YAML specification. Snowflake’s documentation lists the tool types it can expose: Cortex Agents, a coding agent, Cortex Analyst over semantic views, Cortex Search, SQL execution with an optional read-only mode, and custom tools built from UDFs and stored procedures. Each server holds up to 50 tools. OAuth is the recommended sign-in, and programmatic access tokens are available for clients that cannot use OAuth.

## How does access work on Snowflake’s MCP server?

USAGE on the MCP server lets a client list and call tools, and each tool needs its own grant on top: USAGE on a Cortex Agent or search service, SELECT on a semantic view. Sessions use the user’s default role, so Snowflake recommends setting that role to the one meant for MCP access. That works well for a person signing in. An agent that runs on a schedule, or that a whole team talks to in Slack or Notion, has no person behind each call, so it needs a role and a credential of its own. If it borrows someone’s login, everyone who talks to it sees what that person can see.

## Where do Snowflake Cortex and Cortex Analyst fit?

Snowflake Cortex AI is Snowflake’s set of AI features that run inside the account, including AI functions, Cortex Analyst, Cortex Search and Cortex Agents. Cortex Analyst turns a question into SQL over a semantic view, and Cortex Agents plan a sequence of tool calls to answer a question. Both are only as good as the semantic view behind them, and both answer inside Snowflake. Teams that also ask in Claude, Cursor or ChatGPT need the same definitions and the same rules there.

## What about Snowflake Intelligence?

Snowflake’s documentation now describes Snowflake CoWork, the agentic app at the URL that used to cover Snowflake Intelligence. It runs on the Cortex Agent API, uses Cortex Analyst over semantic views for structured data, and inherits Snowflake’s row access policies and column-level security. It suits business users who work in Snowflake’s own app. Rig works alongside it for the people and agents who work in other tools.

## Why do agents on Snowflake need their own role?

An agent is a new kind of user. It can run hundreds of queries a day, it does not get tired of retrying, and it will read whichever table looks closest to the question. On a shared service account it sees more than its task needs, nobody can tell which agent ran which query, and a loop on a schedule keeps a warehouse running. Its own role, a log under its name and a daily call cap deal with each of those in turn.

## How Rig does it on Snowflake

Rig connects to Snowflake with a read-only role and warehouse, reads the information schema, and builds Rig Map over your tables, joins and metric definitions. Gateway then decides who can query what. In Rig, an owner creates an agent from a role they already trust, narrows its schemas, columns and masked fields, sets a daily call cap (1,000 by default) and copies its MCP URL and token into the tool. Before the agent writes SQL it looks up the certified metric and table descriptions in Rig Map, and Rig validates the query before it reaches Snowflake. Context lookups are answered from the map without running a warehouse query, which is where the credit saving comes from. Queries still run as Snowflake SQL on your own compute, and each one lands in the audit log with the agent’s name and tool.

## Who runs agents this way?

Cleo’s data team certifies metrics and table descriptions in Rig, and staff reach them from Claude and Cursor. One Growth Specialist’s agents made 8,023 requests to company data in 30 days, and agent requests at Cleo rose from 6,973 in August to 11,204 in September. Cleo’s warehouse is Redshift. The access model is the same on Snowflake.

## Common questions

What is the Snowflake MCP server?

It is a Snowflake-managed MCP server, created with CREATE MCP SERVER, that exposes Cortex Agents, Cortex Analyst, Cortex Search, SQL execution and custom tools to MCP clients such as Claude and Cursor.

How do I connect Claude to Snowflake with a named agent and a log?

Create the agent in Rig with its own role, token and daily call cap, then paste Rig’s MCP URL and the token into the tool as a bearer credential. Every call the agent makes is logged under its name.

Does Rig replace Cortex or Snowflake Intelligence?

No. Cortex and Snowflake’s own agent app keep working inside Snowflake. Rig serves the same warehouse to Claude, Cursor, ChatGPT and your own agents, with the same definitions and a role for each agent.

How does Rig reduce Snowflake credit burn?

Agents look up tables, joins and certified metrics in Rig Map instead of querying Snowflake to discover them, Rig validates SQL before it runs, and a daily call cap stops a looping agent. We have no published customer figure for the saving, so we model it on your account in an audit.

How do I expose governed business metrics to Claude and Cursor?

Certify the metrics once in Rig Map, then connect Claude or Cursor to Rig’s MCP server. A person signs in with OAuth and gets their own role. An agent gets its own role and token. Either way, the tool looks up the certified metric before it writes SQL, sees only the schemas, columns and rows the role allows, and every call lands in the audit log.

## Related guides and services

[data context and governance](https://rig.so/use-cases/data-context-and-governance)[gateway](https://rig.so/products/gateway)[rig map](https://rig.so/products/rig-map)[mcp](https://rig.so/uses/mcp)[How to create a Snowflake semantic view](https://rig.so/semantic-layer/snowflake)[snowflake context layer and semantic views](https://rig.so/blog/snowflake-context-layer-and-semantic-views)[how to connect ai agents to your data warehouse](https://rig.so/blog/how-to-connect-ai-agents-to-your-data-warehouse)[snowflake](https://rig.so/integrations/snowflake)[cleo](https://rig.so/case-studies/cleo)

## Sources and product details

Vendor information checked on 20 November 2026. Confirm current editions, licence terms and prices with the vendor before buying.

- [Snowflake: Snowflake-managed MCP server](https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-agents-mcp)
- [Snowflake: AI features overview (Cortex AI)](https://docs.snowflake.com/en/guides-overview-ai-features)
- [Snowflake: Cortex Analyst](https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-analyst)
- [Snowflake: Snowflake CoWork (formerly Snowflake Intelligence)](https://docs.snowflake.com/en/user-guide/snowflake-cortex/snowflake-intelligence)
- [Snowflake: semantic views overview](https://docs.snowflake.com/en/user-guide/views-semantic/overview)
- [Model Context Protocol specification](https://modelcontextprotocol.io/specification)
- [Rig guide: give your AI agents their own access](https://rig.so/guides/agent-roles)

Talk to Rig

## What work do you want to improve?

Tell us where your team spends time, loses money or needs better information.
In a 30-minute call, we’ll discuss the data and systems work that could help.

[Book a call with an architect](https://rig.so/book-demo)Use the platform with your team, or add our engineers.

---

- [This page as HTML](https://rig.so/mcp/snowflake)
- [Site map for language models](https://rig.so/llms.txt)
- [API and agent documentation](https://rig.so/developers)
