Setup & Registration

The MCP server is bundled inside the Metatate Native App. Registration creates the canonical MCP server object, grants access to the app roles, and wires the tool layer to your governance tables. The same deployment also creates the core.agent_* wrapper layer used by Snowflake Intelligence custom tools.

Prerequisites

Before registering the MCP server, ensure:

1. Metatate is installed -- The Native App (METATATE_APP) must be installed and the backend service running.
2. APP_WAREHOUSE reference is configured -- The MCP server needs a warehouse to execute procedure-based tools (authorize-use, validate-query-context, explain-why). Bind the APP_WAREHOUSE reference in the Snowflake UI or via SQL:

ALTER APPLICATION METATATE_APP
  SET REFERENCES = ('APP_WAREHOUSE' = '<your_warehouse>');

3. At least one policy is deployed -- The governance tables must contain data. Deploy a policy through the Metatate UI or API before testing the MCP tools.

Registration

Automatic (Recommended)

The MCP server is automatically registered when you run the initialization procedure:

CALL METATATE_APP.v1.init();

init() calls register_mcp_server() internally after configuring schema objects and governance tables. It also runs automatically when the APP_WAREHOUSE reference is first configured.

Manual

If you need to re-register the MCP server (e.g., after a schema change or troubleshooting), call it directly:

CALL METATATE_APP.v1.register_mcp_server();

This is idempotent -- calling it multiple times is safe. It will drop and recreate the MCP server object if it already exists.

What Registration Creates

Registration creates the following objects in the core schema:

Object	Type	Description
core.metatate_mcp	MCP Server	Canonical Metatate machine interface
core.discover_context	Function	Searches governed tables with optional filters
core.get_decision_context	Function	Returns full governance context for a table
core.inspect_data_meaning	Function	Returns column-level governance detail
core.authorize_use	Procedure	Evaluates an authorization request against policies
core.validate_query_context	Procedure	Analyzes SQL for governance compliance
core.explain_why	Procedure	Retrieves audit trail for a past decision
core.agent_*	Functions / Procedures	Snowflake Intelligence-friendly wrappers with scalar parameters

Grants

Registration grants USAGE on the MCP server and USAGE/EXECUTE on the underlying functions and procedures to both app roles:

• app_admin -- Full access to all tools
• app_user -- Full access to all tools (governance is data-driven, not role-gated at the tool level)

Verifying Registration

Check the MCP Server Exists

USE APPLICATION METATATE_APP;
SHOW MCP SERVERS IN SCHEMA CORE;

Expected output: one row for METATATE_MCP.

Check the Tools

SHOW FUNCTIONS IN SCHEMA METATATE_APP.CORE;
SHOW PROCEDURES IN SCHEMA METATATE_APP.CORE;

You should see the six tool objects listed in the table above.

Test a Tool Directly

The fastest way to verify the MCP server is working is to call a function directly:

-- Discover all governed tables (no filters)
SELECT METATATE_APP.CORE.DISCOVER_CONTEXT(
  OBJECT_CONSTRUCT()
);

Expected: A JSON object with status: "success" and a data array containing your governed tables.

-- Get governance context for a specific table
SELECT METATATE_APP.CORE.GET_DECISION_CONTEXT(
  OBJECT_CONSTRUCT('table_name', 'MY_DB.MY_SCHEMA.CUSTOMERS')
);

Test the Snowflake Intelligence wrapper pattern

You can validate the Snowflake Intelligence adapter layer directly in SQL before wiring it into an agent:

SELECT METATATE_APP.CORE.AGENT_GET_DECISION_CONTEXT('MY_DB.MY_SCHEMA.CUSTOMERS');

This should return the same JSON payload as the canonical GET_DECISION_CONTEXT tool.

In-App Testing

Metatate includes a Test Tools tab in the application UI. This provides a graphical interface to:

• Browse all six tools and their input schemas
• Fill in parameters with form fields
• Execute tools and view formatted JSON responses
• Compare canonical tool payloads with the Snowflake Intelligence wrapper behavior

This is the easiest way to explore the canonical MCP layer without writing SQL or configuring a Snowflake Intelligence agent first.

Wrapper regression script

To validate the full Snowflake Intelligence adapter layer against a deployed app, run:

snow sql -f scripts/test_snowflake_intelligence_tools.sql

That script seeds acceptance-style governance fixtures, executes all six core.agent_* wrappers, and cleans up the test data afterward.

Unregistering

To remove the MCP server (e.g., before uninstalling):

USE APPLICATION METATATE_APP;
DROP MCP SERVER IF EXISTS CORE.METATATE_MCP;

The underlying functions and procedures will remain. Re-run register_mcp_server() to recreate the server object.

Next Steps

• Tool Reference -- Detailed documentation for each canonical tool
• Snowflake Intelligence -- Build an Intelligence agent using the core.agent_* wrappers