SingleStore MCP

The SingleStore MCP server exposes both the data plane and the control plane of SingleStore Helios to an AI client. run_sql executes queries against a connected workspace and returns structured results, which covers exploration, schema questions and ad hoc analysis. Around it sit workspace tools: workspace_groups_info and workspaces_info describe the deployed infrastructure, resume_workspace brings a suspended workspace back, and list_regions and list_sharedtier_regions enumerate where new deployments can be placed. Starter workspaces have their own lifecycle tools to list, create and terminate them, so an agent can spin up a scratch environment and clean it up afterwards. Notebook and job tools extend the server past querying. create_notebook_file and upload_notebook_file put a notebook into the workspace, create_job_from_notebook schedules that notebook to run on a recurring basis, and get_job and delete_job monitor and remove scheduled jobs. That combination lets an assistant go from an exploratory query to a scheduled pipeline in one session. Account context tools get_user_info, organization_info and set_organization keep multi organization users pointed at the right tenant. The server runs as a Python package launched with uvx, and an init command wires it into a chosen client. Authentication normally goes through a browser based OAuth flow, with API keys used for container deployments. Supported clients include Claude Desktop, VS Code, Cursor, Windsurf, Gemini CLI, LM Studio and Goose. It targets database administrators and developers working on SingleStore.