What it does
Warehouse Context is an MCP server that gives agents documented context about a data warehouse: what each table is for and who owns it, its columns and PII flags, upstream sources and downstream consumers, how fresh it is, and a few sample rows with PII masked. It offers a single run_query tool for one read-only SELECT. It supports DuckDB, Snowflake and Databricks, and merges comments and tags stored in the warehouse with a curated YAML catalog, where curated values take precedence.
An agent would reach for it when it needs to know which table to trust before writing SQL, for example to find the governed source for a metric, confirm a table has been updated recently, or avoid exposing personal data. Queries pass through several independent layers: an allow-list parser (sqlglot) that accepts only one SELECT built from reviewed constructs, checks that every column resolves to the real schema, masks PII columns inside the SQL so values never leave the warehouse, uses a read-only or SELECT-only connection, and enforces row, time and size limits. Attack queries from an independent review are kept in the test suite and run on every commit.
To try it, the server can run with a bundled synthetic demo warehouse through uvx. The repository also includes configuration for Claude Code, Claude Desktop, VS Code and Codex, a Docker option that serves HTTP, and a Claude Code plugin with a warehouse-context skill. Connecting a real warehouse means switching to the Snowflake or Databricks backend and supplying its settings.
Limitations and requirements
Requires uv to run. A real warehouse needs its own backend settings. The DuckDB backend (v0.1) reads no native warehouse context, only the YAML catalog; Snowflake and Databricks are v0.2. PII tags must be readable at startup or the server fails closed, and they are refreshed every 10 minutes. The publisher states that warehouse-native masking (masking policies or column masks) should also be used for production data, since the server's masking is defence in depth. Development requires Python 3.14, which uv installs if missing.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
Warehouse Context
Governed warehouse context for agents: docs, lineage, freshness and read-only, PII-safe SQL.
v0.5.1As shown on registry.modelcontextprotocol.io when we last read it