What is Metatate Cloud?

Metatate Cloud turns the passive metadata in your data estate — policies, classifications, meaning, and usage rules — into an operational decision layer that AI agents and humans can query at the point of use. It works across multiple data platforms (PostgreSQL, MySQL, BigQuery, Redshift, Databricks, and Snowflake), reads schema metadata instead of copying production data, and answers one recurring question with authority: what does this data mean, and how may it be used?

From metadata to typed answers

Everything Metatate Cloud serves flows through one governed pipeline:

1. Catalog vocabulary. Connectors sync schema metadata into a catalog you curate: descriptions, classifications against a sensitivity taxonomy, collections, and business meaning.
2. Canonical policies. You author policies as documents that are reviewed and approved. Each approved version is a canonical policy — versioned, diffable, and auditable.
3. Deployable instructions. The governance engine derives deployable instructions from your canonical policies and the catalog assets they target.
4. Deployment review. Changes are packaged into a deployment plan that you review before anything goes live.
5. Current deployment publication. Publishing the plan produces the current deployment publication, which carries the deployed instruction decisions — the exact rows every decision-bearing answer reads.
6. Typed answers. Agents query the publication over MCP and get a typed answer in one of three states: answered, review_required, or not_enough_published_state. When published state is missing, unresolved, conflicted, or outside its validity window, the answer says so — it never fabricates a decision.
7. A way forward when the answer is no. A blocked agent can file a review request for a human steward, who may grant a scoped, expiring exception. The agent retries; the server verifies the exception before anything is satisfied.

The MCP tools describe decisions and cite the policy versions behind them. They never enforce, mask, or block anything in your systems, and they never change your governance — the single exception is a narrow, honestly advertised write lane that files a review request for a human steward.

What you can do

• Connect data sources. Sync catalog metadata from the platforms where your data lives. See the data sources overview.
• Author policies. Write governance policies with AI assistance, then submit them for review and approval.
• Review and publish. Deploy approved policies through a reviewed plan into the current deployment publication.
• Serve answers to agents. Expose nine MCP tools — discovery, meaning, rules, purpose-aware authorization, query validation, decision explanation, and the steward request lane — to any MCP-capable agent. See the MCP overview.

Where to go next

• Quick start — from sign-up to your first governed answer in about five minutes.
• Create a workspace — sign up, set up your workspace, and invite your team.
• Load the Customer 360 sample — a one-click synthetic estate with approved policies and a live publication.
• Data sources overview — supported platforms and how catalog sync works.
• MCP overview — the tool surface agents see.
• Governance model — how policies become published, queryable decisions.
• Purpose keys — stating why data is used, and when that changes the answer.
• Examples — runnable notebooks over the Customer 360 estate, offline or live.

Ready to try it? Create a free account and start on the free plan. Paid tiers are described on the plans page.