Validatar Agent Kit for Snowflake Cortex Code

Set up the Validatar Agent Kit in Snowflake Cortex Code (the cortex CLI and Cortex Code Desktop). When you finish, Cortex Code can explore your Validatar catalog, create and run tests against your Snowflake data, and package test packs.

Note: Cortex Code in Snowsight connects only to MCP servers that use OAuth, so it cannot use the Validatar MCP server today. Use the CLI or the Desktop app.

Before you start

• Cortex Code CLI or Desktop installed and connected to your Snowflake account.
• Python 3.9 or newer on the PATH (python --version).
• Your Validatar instance address, for example https://acme.cloud.validatar.com.
• A Validatar user that can create tests in at least one project, ideally a sandbox project.

Get the kit

1. Download Validatar Agent Kit for Snowflake Cortex Code from the Validatar Marketplace.
2. Extract the zip. It holds one folder, validatar-kit-cortex-code/. Use it as your working folder, or copy its contents into an existing repository (see below).

Create your tokens and set the variables

1. In Validatar, open your user profile, User Tokens. Create a token of type MCP. For MCP + API mode (optional), also create a token of type API with the scopes Authoring, Execution, Data Steward and Catalog.

2. Set the environment variables in the shell you start cortex from (Cortex Code reads the environment of the process that launched it):

   Variable	Value
   VALIDATAR_URL	Your instance address, no trailing slash
   VALIDATAR_MCP_TOKEN	The MCP token
   VALIDATAR_API_TOKEN	The API token (optional: MCP + API mode)

   Windows (PowerShell): setx VALIDATAR_URL "https://acme.cloud.validatar.com", and the same for each token, then open a new terminal. macOS or Linux: export lines in ~/.zshrc or ~/.bashrc.

Add the MCP server

Cortex Code keeps MCP servers in your user profile, not in the project. In the kit folder, run:

python validatar/scripts/setup.py configure --user

It adds this entry to ~/.snowflake/cortex/mcp.json and backs up the file first:

{
  "mcpServers": {
    "validatar": {
      "type": "http",
      "url": "https://acme.cloud.validatar.com/core/mcp",
      "headers": { "x-val-api-token": "${VALIDATAR_MCP_TOKEN}" }
    }
  }
}

The header reads the token from the environment when Cortex Code starts. Alternatively, run cortex mcp add validatar https://acme.cloud.validatar.com/core/mcp --transport http -H "x-val-api-token: <token>" in your own terminal; Cortex Code moves the header value to your operating system's keychain.

Then run python validatar/scripts/setup.py check.

Start Cortex Code

1. In a terminal where the variables are set, go to the kit folder and run cortex.
2. Run /mcp: the validatar server is listed. Run /skill list: the validatar-* skills from .cortex/skills/ are listed.
3. Ask: "Connect to Validatar and tell me what projects and data sources I have."

The always-on instructions are in AGENTS.md.

Limits to know

• MCP calls time out after 60 seconds by default. For long queries or runs, set COCO_MCP_TOOL_TIMEOUT_MS=300000 before starting cortex.
• Tool results over 50 KB are cut off. The kit's skills page large lists and ask for specific sections, but ask for smaller pieces if an answer looks incomplete.

Use the kit in an existing project

Copy AGENTS.md (or append its content to yours), .cortex/skills/validatar-*, the validatar/ folder and the lines from .gitignore. The MCP server entry is already in your user profile.

Troubleshooting