data serve tool (dst)
An open-source semantic layer and MCP server that sits between the AI tools a team uses and its data warehouse, answering questions in plain language with the SQL that produced the answer, a confidence grade, and a receipt.
dst lets an AI assistant such as Claude, ChatGPT, or Copilot ask a data question in plain language and get an answer resolved from definitions the team wrote down, rather than a guess. Each answer returns with the SQL that produced it, a confidence grade, and a receipt, and an ambiguous term produces a clarifying question or a refusal instead of a wrong number.
It sits in the data stack after loading and transformation -- dlt loads, dbt transforms, dst serves -- and connects to Snowflake, BigQuery, Postgres, or DuckDB over MCP. It runs with a bundled demo warehouse, its own Postgres instance in Docker, and a model provider such as Anthropic or any OpenAI-compatible endpoint.
The server exposes a dashboard alongside its MCP endpoint for inspecting the ledger of answers it has served.