MCP server
Point external AI clients at Sundial over MCP to ask analysis and modelling questions with your governed context.
Sundial exposes a Model Context Protocol (MCP) server, so AI clients you already use, Claude, Cursor, and other MCP-capable tools, can query Sundial directly. Instead of copying data around, your client asks Sundial, and Sundial answers using your governed semantic layer, playbooks, and context.
What it exposes
Connecting to Sundial's MCP server gives a client two tools:
ask_sundial_analyst: ask analytical questions (the same capability as the Analysis Agent), with follow-ups.ask_sundial_modeler: ask the Data Modelling Agent about, and make changes to, your semantic layer and models.
Because these run inside Sundial, answers keep the same grounding and trust signals you get in the app; your external client gets governed analytics, not a raw warehouse pipe.
Connecting
The MCP server uses OAuth: your client initiates a connection, you authorize it against Sundial, and the client can then call the tools on your behalf. Point your client at Sundial's MCP endpoint and complete the OAuth prompt; your Sundial contact can provide the endpoint URL for your deployment.
When to use this
Use the MCP server when you want Sundial's answers inside another agent or workflow, for example asking Sundial from your coding assistant, or composing Sundial into a larger multi-tool agent. To work the other way, letting Sundial read from your other tools, see Tool connectors.