Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.spojit.com/llms.txt

Use this file to discover all available pages before exploring further.

The Snowflake connector lets your workflows talk to Snowflake’s managed MCP server — a remote MCP endpoint hosted by Snowflake that exposes Cortex Analyst (text-to-SQL), Cortex Search (semantic search over your data), Cortex Agents, SQL orchestration, and object management as tools, with built-in role-based access control.
Spojit does not curate the tool list — Snowflake does. For the live set of tools and their parameters, see Snowflake’s MCP server documentation.

​
Before you connect

Snowflake’s managed MCP requires account-side setup before the connector can reach it. A Snowflake admin needs to:
1

Create an MCP server object in Snowflake

Run CREATE MCP SERVER in a worksheet. The MCP server object defines which Cortex agents, Cortex Search services, and SQL endpoints are exposed via the MCP endpoint. Choose the database, schema, and name — you’ll enter these values in the Spojit connection form below.See Snowflake’s documentation on creating MCP servers for the full DDL syntax.
2

Grant role access

Grant USAGE on the MCP server object to the role(s) that should be able to invoke it. The user whose PAT you’ll use in Spojit must have one of those roles set as the default.
3

Create a Programmatic Access Token (PAT) for the user

In the Snowflake UI, open the user’s profile → Programmatic Access Tokens → Generate token. Copy the token (Snowflake only shows it once). Scope it to a role that has USAGE on the MCP server.

​
Connection setup

1

Add the connection in Spojit

Go to Connections in Spojit, click Add Connection, and select Snowflake (Official MCP).
2

Fill in the fields

  • Account URL — your Snowflake account URL without protocol or trailing slash (e.g. xy12345.us-east-1.snowflakecomputing.com). Find it in your Snowflake URL bar after logging in.
  • Database — the database that contains the MCP server object you created (e.g. MY_DB).
  • Schema — the schema that contains the MCP server object (e.g. PUBLIC).
  • MCP Server Name — the name you gave the MCP server when you ran CREATE MCP SERVER (e.g. MY_MCP).
  • Programmatic Access Token — paste the PAT you generated in step 3 above.
3

Test the connection

Click Test connection or drag the connector into a workflow — the tool list should populate with whatever Cortex agents and SQL endpoints your MCP server object exposes.

​
Caveats

  • Account-side setup is the long pole. The MCP server object, the role grants, and the PAT all need to exist in Snowflake before the Spojit connection works. If tools/list comes back empty or you get a 404, the most likely cause is a mismatch between the database/schema/name you entered and the MCP server object’s actual location.
  • The PAT carries the user’s role permissions. A PAT can do anything the underlying user role can do. Scope the role narrowly — read-only on tables the agent needs to query, no OWNERSHIP privileges, etc. Treat the PAT itself like a password.
  • Tool surface is determined by your MCP server object. Snowflake’s managed MCP doesn’t expose every Snowflake API — it exposes whatever you registered when you ran CREATE MCP SERVER. To add or remove tools, edit the MCP server object in Snowflake; no Spojit-side change needed.
  • MCP endpoint is hosted by Snowflake. Availability, rate limits, and feature evolution are upstream of Spojit. The endpoint URL is per-account (https://<account>.snowflakecomputing.com/api/v2/databases/.../mcp-servers/...) and resolved at request time from your connection fields.

​
Tools

For the current list of tools and their parameters, see Snowflake’s MCP server documentation. The exact tool set depends on what you registered in your CREATE MCP SERVER statement.