Vanna AI

You can use the Vanna AI MCP Server to translate natural language questions into SQL queries against a remote database, securely connected through SSH. This enables AI assistants to query your data sources via MCP clients, returning results or running queries directly through an integrated workflow.

How to use

Connect your MCP client (such as Claude Desktop or Gemini CLI) to the Vanna AI MCP Server using one of the available methods. You can run the server locally and expose it over HTTP, or run it as a local process that the MCP client communicates with via stdio.

How to use with an MCP client

• Start the server locally (see installation steps) and note the HTTP endpoint or the local stdio command your client can use.

What you can do with it

• Ask questions in plain English and have the server generate SQL queries automatically. - Execute the generated SQL against your remote database through a secure SSH tunnel. - Train Vanna on your database schema and sample queries to improve precision over time. - Persist trained data in a vector store for fast retrieval and pattern reuse.

Training and data management

• Use the provided tools to train Vanna with example questions and corresponding SQL, or train on your database schema (DDL) to better reflect your data structure. - View and manage your trained examples to refine responses.

Security and access

• SSH tunneling ensures your database credentials are transmitted over a secure channel. - Prefer SSH key authentication when possible and keep credentials out of reach. - Keep your environment file secure and never commit it to version control.

Available tools

generate_sql

Generate SQL from a natural language question to fetch the desired data.

run_sql

Execute a SQL query and return the results.

train_question_sql

Train Vanna with example question and corresponding SQL pairs to improve translation accuracy.

train_ddl

Train Vanna on database schema definitions (DDL statements) to understand structure.

get_training_data

View all trained examples and patterns stored for future queries.