Your AI tools are powerful.
Your data is locked behind walls they can't see.
Rig MCP gives Codex, Claude Code, Cursor, ChatGPT, and any AI tool governed access to your warehouse — schemas, metrics, joins, and live data through one protocol.
See Rig MCP in action
Watch an AI coding agent discover tables, resolve joins, and query a warehouse — all through Rig MCP.
One context layer, every AI tool
Any tool that supports MCP can discover your tables, understand your schema, and query your warehouse — without ever seeing raw credentials.
How Rig MCP works
"Find tables related to subscription revenue and show me the join path to users"
Schemas, column descriptions, join keys & sample data
Governed, read-only, auto-limited — no raw warehouse credentials ever leave Rig
Built for safe, governed access
OAuth
Every user's data access is managed via OAuth. Set and audit permissions and logs per user and per team.
Pre-computed context
Metadata, embeddings, join graphs, and sample values served from an in-process database — no runtime LLM calls for search.
Sandbox-first SQL
All agent queries validated in a sandbox. Execution auto-limits and respects permissions.
Common questions
An MCP (Model Context Protocol) server that gives AI tools, like Codex, Claude Code, Cursor, ChatGPT, and Claude Cowork, governed access to your warehouse: schemas, metrics, joins, and live data through one protocol, without ever exposing raw database credentials.
Any tool that supports the Model Context Protocol. Rig exposes 17 MCP tools for semantic table search, join and lineage resolution, and sandboxed querying, so the agent gets exactly the context it needs and nothing more.
Role-based access control is enforced on every query, all SQL is validated in a sandbox in under 10ms and auto-limited, access is managed via OAuth, and raw credentials never leave Rig. Every request is permissioned and logged for audit.
No. Context such as metadata, embeddings, join graphs, and sample values is pre-computed and served from an in-process database, cutting warehouse compute overhead by around 90%, with no runtime LLM calls needed for search.