/snowflake-cortex

name: snowflake-cortex
description: >-
  Use Cortex AI services — Cortex Analyst for natural-language-to-SQL
  through semantic views, Cortex Search for semantic search, and Cortex
  Agents for orchestrated AI tasks.

Snowflake Cortex AI

Use Snowflake's Cortex AI services when they are available on the MCP server.

When to use

• User asks a data question and semantic views are configured (Cortex Analyst)
• User wants to search unstructured data like support tickets or documents (Cortex Search)
• User wants to interact with a Cortex Agent for multi-step tasks

Discover available Cortex services

-- Semantic views (data models for Cortex Analyst)
SHOW SEMANTIC VIEWS IN ACCOUNT;

-- Cortex Search services
SHOW CORTEX SEARCH SERVICES IN ACCOUNT;

-- Cortex Agents
SHOW AGENTS IN ACCOUNT;

Cortex Analyst

Cortex Analyst converts natural language questions into SQL using semantic views. If the MCP server exposes a `CORTEX_ANALYST_MESSAGE` tool, pass the user's question as the `message`.

**When to prefer Analyst over raw SQL:**

• The question maps to metrics/dimensions defined in a semantic view
• The user is asking business questions ("what was revenue last quarter?")
• Governance and auditability matter

**When to use raw SQL instead:**

• Admin tasks (SHOW, DESCRIBE, GRANT)
• DDL/DML operations
• No semantic view covers the domain

Cortex Search

Cortex Search provides semantic search over unstructured text columns. If the MCP server exposes a `CORTEX_SEARCH_SERVICE_QUERY` tool:

{
  "query": "billing dispute resolution",
  "columns": ["TICKET_ID", "CUSTOMER_NAME", "REQUEST"],
  "limit": 10
}

**Filter syntax** (optional):

{
  "query": "network outage",
  "filter": { "@eq": { "SERVICE_TYPE": "Cloud" } },
  "limit": 5
}

Supported filter operators: `@eq`, `@contains` (arrays), `@gte`, `@lte`, composed with `@and`, `@or`, `@not`.

Cortex Agents

If the MCP server wraps a Cortex Agent (`CORTEX_AGENT_RUN`), send a `text` message. The agent decides which internal tools to use (SQL, Analyst, Search, code execution).

For multi-turn conversations, include `thread_id` from the previous response.

Examples

• "What was total revenue by region last quarter?" → Route to Cortex Analyst
• "Search support tickets about billing errors" → Route to Cortex Search
• "Analyze our customer churn data and create a visualization" → Route to Cortex Agent