The failure mode of natural-language SQL is a model guessing at table and column names, which is what `get_schema_context` exists to prevent — it retrieves the schema that matches your question before anything is written. Saved use cases turn the good queries into ones the team can re-run rather than re-derive.
A hosted MCP server at https://ai.quanti.io/mcp that puts your Quanti projects and their warehouse schema in front of an assistant. It builds schema context with RAG, then your assistant writes the SQL and the server executes it read-only.
- list_projects — the Quanti projects you can reach
- get_project_context — connectors, datasets and date ranges for a project
- get_schema_context — schema context built for your specific question via RAG
- execute_query — run a read-only SQL query against the warehouse
- get_use_cases — search the library of use case templates
- list_my_use_cases / list_project_use_cases — your saved analyses and the team's shared ones
- create_use_case / update_use_case / delete_use_case — save an analysis for reuse and maintain it
- collect_feedback — send feedback on a query result
A Quanti account with access to at least one project, a warehouse with an indexed schema (BigQuery, Snowflake, Redshift or Azure Synapse), and a client that supports MCP. Authentication is OAuth 2.0 with dynamic discovery; transport is HTTP with SSE. Setup guides ship for ChatGPT, Claude Desktop, Gemini, Le Chat and Copilot. Queries are read-only and access is controlled per project.