Connect assistants to Cloud

The query MCP lets an assistant discover your semantic layer and ask governed questions. Connect an MCP client that supports remote Streamable HTTP.

Query MCP with an API key

Use this workspace and environment URL, replacing both placeholders:

https://cloud.semantic-rails.com/v1/tenants/<workspace id>/runtime/<env>/mcp

Choose dev, staging or prod. Use an API key bound to the same environment with query:execute. Add query:sql if the assistant should be allowed to see SQL.

Send the key with the ApiKey scheme: Authorization: ApiKey <your key>. Cloud reads Bearer as an OAuth sign-in token, so an API key sent as Bearer is refused with 401. When the environment serves more than one project, also send X-Semantic-Package: <project name>.

Workspace setup shows the URL and a ready-made client configuration with a key placeholder. Copy or download it, add your key privately in the client's settings, and select a project when prompted. An owner or admin can provide the environment key.

The selected environment needs an active project version and a bound warehouse. Preparing a configuration does not verify the assistant connection. Ask the assistant to discover the available metrics after connecting.

Limits

Query calls are limited per workspace, per minute, by plan. MCP and the REST runtime API share the same limits. On the trial and Builder plans, a workspace can make 6 calls a minute that validate, compile or run a query (every MCP execute call), and 60 discovery and planning calls a minute. Higher plans allow more. Over a limit, Cloud answers HTTP 429 with code rate.limited; retry after the number of seconds in details.retry_after_seconds.

For the sample engine endpoint, see the hosted MCP demo. For your own Cloud data, connect a warehouse and create and promote a project.