Your schema says what the data is. Metatate's MCP server tells the agent what the data means, who may use it, and for what — before the query runs, over the protocol every agent already speaks.
The same seven tools, the same typed answers — whether the server runs inside your Snowflake account or alongside the rest of your stack.
Runs as a Snowflake Native App inside your account. Cortex Agents and Cortex Code call the tools over MCP or as native SQL functions. Your data and your policies never leave Snowflake.
Metatate for Snowflake →A hosted MCP endpoint for Postgres, BigQuery, Redshift, Databricks, and Snowflake estates. Each environment gets its own scoped endpoint; agents authenticate with tokens you issue and revoke from the app.
Metatate Platform →Discovery, meaning, rules, authorization, validation, and explanation — the questions an experienced engineer asks before touching a table, exposed as MCP tools.
discover_context
Find governed assets by database, schema, domain, sensitivity, or compliance tag.
get_decision_context
Pull the full policy and meaning bundle for a specific asset before deciding.
inspect_data_meaning
Read column-level meaning, classification, sensitivity, PII, and masking facts.
inspect_governance_rules
List active usage, validation, and transfer rules for a governed asset.
authorize_use
Resolve allow / conditional / deny for an intended use, with rationale and a decision id.
validate_query_context
Check a SQL or query plan against active rules before it runs.
explain_why
Return the policy, version, and rule that produced any prior decision.
The tools answer questions about published governance state. They never approve, publish, mask, block, or write it. Every read is pinned to the tenant the token belongs to.
answered
Published rules are sufficient. The agent gets a decision it can act on.
review_required
Rules exist but conflict or explicitly require a human. The agent knows to stop.
not_enough_published_state
Governance hasn't covered this case. The agent knows it's guessing — before it guesses.
Typed answers with stable reason codes mean an agent can tell a real decision from a gap in coverage — and knows when the right move is to ask a human instead of improvising.
The Metatate plugin adds eight slash commands so Claude reads your data's policies before it writes a query.
Metatate for Claude →The Cortex Code plugin brings the same decision layer to Snowflake's coding agent, in-account.
Plugin on GitHub →MCP over HTTP with bearer tokens. If your agent framework speaks MCP, it can check the rules before it acts.
Connection guide →Connect a database, author policies in plain language, and point any MCP client at your endpoint. Free tier, no card.
Start free