Getting Started with Managed Snowflake MCP Server

The Snowflake MCP Server allows AI agents to securely retrieve data from Snowflake accounts without needing to deploy separate infrastructure. MCP clients discover and invoke tools, and retrieve data required for the application. The Snowflake MCP Server includes Cortex Analyst and Cortex Search as tools on the standards-based interface. It is now available with Model Context Protocol (MCP) so that AI Agents can discover and invoke tools (Cortex Analyst, Cortex Search) via a unified and standard based interface.

Top 3 Benefits

Governed By Design: Enforce the same trusted governance policies, from role-based access to masking, for the MCP server as you do for your data.
Reduced Integration: With the MCP Server, integration happens once. Any compatible agent can then connect without new development, accelerating adoption and reducing maintenance costs.
Extensible Framework: Provide agents with out-of-the-box secure access to structured and unstructured data. You can refine the tools to improve how agents interact with your data.

Why It Matters

MCP Server on Snowflake simplifies the application architecture and eliminates the need for custom integrations. Enterprises can expedite delivery of generative AI applications with richer insights on a standards based architecture and a robust governance model with the Snowflake AI data cloud.

NOTE: Snowflake MCP Server is in Public Preview as of October 2025.

MCP

Prerequisites

Access to a Snowflake account with ACCOUNTADMIN role. If you do not have access to an account, create a free Snowflake trial account.
Access to Cursor.

What You Will Learn

How to create building blocks for Snowflake MCP Server that can intelligently respond to questions by reasoning over data
How to configure Cursor to interact with Snowflake MCP Server

What You Will Build

A Snowflake MCP Server that intelligently responds to questions by reasoning over data from within Cursor.

Create Objects

Clone GitHub repo.
In Snowsight, create a SQL Worksheet and open setup.sql to execute all statements in order from top to bottom.

Create Personal Access Token

Create a Personal Access Token (PAT) for your role and make a note/local copy of it. (You will need to paste it later.)

Create Cortex Search Service

This tool allows the agent to search and retrieve information from unstructured text data, such as customer support tickets, Slack conversations, or contracts. It leverages Cortex Search to index and query these text "chunks," enabling the agent to perform Retrieval Augmented Generation (RAG).

In Snowsight, on the left hand navigation menu, select AI & ML » Cortex Search
On the top right, click on Create
- Role and Warehouse: ACCOUNTADMIN | DASH_WH_S
- Database and Schema: DASH_MCP_DB.DATA
- Name: Support_Tickets
- Select data to be indexed: select FACT_SUPPORT_TICKETS table
- Select a search column: select DESCRIPTION
- Select attribute column(s): select CATEGORY, SUBCATEGORY, PRIORITY, CHANNEL, STATUS, SATISFACTION_SCORE
- Select columns to include in the service: Select all
- Configure your Search Service: Keep default values and select DASH_WH_S for "Warehouse for indexing"

Create Snowflake MCP Server

To create the Snowflake MCP server, run the following in the same SQL worksheet.

create or replace mcp server dash_mcp_server from specification
$$
tools:
  - name: "Support Tickets Search Service"
    identifier: "dash_mcp_db.data.support_tickets"
    type: "CORTEX_SEARCH_SERVICE_QUERY"
    description: "A tool that performs keyword and vector search over support tickets and call transcripts."
    title: "Support Tickets"
$$;

Now let's try this out in Cursor, but note that you should be able to use other clients like CrewAI, Claude by Anthropic, Devin by Cognition, and Agentforce by Salesforce.

Cursor

In Cursor, open or create mcp.json located at the root of your project and add the following. NOTE: Replace and with your values.

{
    "mcpServers": {
      "Snowflake": {
        "url": "https://<YOUR-ORG-YOUR-ACCOUNT>.snowflakecomputing.com/api/v2/databases/dash_mcp_db/schemas/data/mcp-servers/dash_mcp_server",
            "headers": {
              "Authorization": "Bearer <YOUR-PAT-TOKEN>"
            }
      }
    }
}

Then, select Cursor -> Settings -> Cursor Settings -> MCP (or Tools & MCP) and you should see Snowflake under Installed Servers.

NOTE: If it continues to say "Loading tools" running the following curl command to test your connection.

curl -X POST "https://<YOUR-ORG-YOUR-ACCOUNT>.snowflakecomputing.com/api/v2/databases/dash_mcp_db/schemas/data/mcp-servers/dash_mcp_server" \
  --header 'Content-Type: application/json' \
  --header 'Accept: application/json' \
  --header "Authorization: Bearer <YOUR-PAT-TOKEN>" \
  --data '{
    "jsonrpc": "2.0",
    "id": 12345,
    "method": "tools/list",
    "params": {}
  }'

Q&A

Assuming you're able to see the tool Support_Tickets_Search_Service under newly installed Snowflake MCP server, let's chat! Start a new chat in Cursor and set your mcp.json as context to ask the following questions.

Q1. Can you summarize the overall sentiments based on the support calls?

In a few moments, you should see a response similar to this.

NOTE: The response may not be the exact same.

Based on my analysis of the support tickets data, here's a comprehensive summary of the overall sentiments from support calls:
Overall Sentiment Summary

The support call data reveals three distinct sentiment categories with varying customer experiences:

🟢 Positive Sentiments (~35-40% of feedback)
Key Characteristics:
High praise for agent professionalism: Customers specifically commend agents who are "professional, knowledgeable, and genuinely cared about helping"
Problem resolution satisfaction: Quick resolution of complex issues (e.g., account balance discrepancies, wire transfer problems)
Communication excellence: Agents keeping customers informed throughout the process
Persistence and dedication: Support staff staying on calls until complete resolution
Representative feedback:
"Outstanding customer care!"
"They investigated thoroughly and found the missing deposit"
"The agent stayed on the line until everything was resolved"

🔴 Negative Sentiments (~30-35% of feedback)
Major Pain Points:
Long wait times: Customers reporting being on hold for over an hour
Multiple transfers: Frustration with being transferred 3+ times without resolution
Agent unprofessionalism: Reports of rude, dismissive, or annoyed agents
Account issues: Unexpected account freezes causing public embarrassment
Poor communication: Agents unable to explain problems or rushing through explanations
Representative feedback:
"Completely unprofessional"
"I've been transferred three times already and still no resolution"
"The embarrassment was terrible and the agent couldn't explain why it happened"

🟡 Neutral/Informational (~25-30% of feedback)
Characteristics:
Routine inquiries: Questions about interest rates, account tiers, balance requirements
Straightforward transactions: Basic information requests handled efficiently
No strong emotional response: Professional but unremarkable interactions
Representative feedback:
"Everything was straightforward"
"The agent provided the information and explained the different account tiers"

Key Insights
Bimodal Distribution: Customer sentiment tends to be either very positive or very negative, with fewer neutral experiences
Agent Training Impact: The difference between positive and negative experiences often comes down to individual agent skills and approach
Process Issues: System problems (account access, multiple transfers) contribute significantly to negative sentiment
Resolution Speed: Quick, thorough problem-solving generates the strongest positive feedback

Recommendations
Focus on reducing wait times and transfer frequency
Enhance agent training on communication and problem-solving skills
Improve system reliability to prevent unexpected account issues
Standardize the excellent service that generates positive feedback across all interactions

The data suggests a support organization with high potential but inconsistent execution, where exceptional service exists alongside significant operational challenges.

Q2. Which support categories would benefit most from automated responses based on transcript analysis?