Rudder Lookout is currently in Public Beta. Contact the Product team if you have any questions.
Lookout exposes a curated set of its tools to external AI tools through the Model Context Protocol (MCP). This lets an approved external assistant or agent call into your workspace programmatically, for example, to run a warehouse query or read a dashboard, using the same connections your team already set up.
External tools get a curated, mostly read-only subset of Lookout’s capabilities. For the exact, current tool list, see the in-app MCP reference in your workspace.
| Tool | Permissions | Description |
|---|---|---|
| Warehouse | Read | Query a connected warehouse, list tables, describe a table, and list warehouses |
| Dashboards | Read | List dashboards, get a dashboard, and run a single dashboard panel |
| Source documentation | Read | List and read the documentation Lookout generated for your sources. This is the generated documentation in the Context Hub, not your RudderStack source catalog |
| Dashboards | Write | save_dashboard creates or updates a dashboard. It is the only write tool and is gated behind the mcp:write scope (see below) |
When an external tool authorizes via OAuth, or when you create a named programmatic token, the default scope is mcp:read mcp:write, which includes the save_dashboard write capability.
To grant read-only access, explicitly request the mcp:read scope only.
Because the default scope includesmcp:write, an external tool that doesn’t narrow its scope can save dashboards in your workspace. If a tool only needs to read, grant itmcp:readonly.
Any member can create clients and tokens for external access from the MCP area in the sidebar (also reachable from Settings > Integrations). External tools then call into the workspace using those credentials.
External AI tool access is enabled at the deployment level, not per workspace. If you don’t see the MCP area, it may not be enabled for your deployment — check with your admin.