Install Vanna AI MCP Server

Pick your client, copy the command, done.

claude mcp add -e "LLM_TYPE=${LLM_TYPE}" vanna-mcp -- uv run main.py

Required:LLM_TYPE+ 3 optional

Environment Variables

Set these before running Vanna AI MCP Server.

VariableDescriptionRequired

LLM_TYPEThe LLM provider to use (ollama, lmstudio, claude, gemini, openai)Yes

ANTHROPIC_API_KEYAPI key for ClaudeNo

GEMINI_API_KEYAPI key for GeminiNo

OPENAI_API_KEYAPI key for OpenAINo

Once configured, Vanna AI MCP Server gives your AI agent access to:

ask_databaseConvert a question to SQL and optionally execute it.

question

train_vannaProvide DDL or example SQL to teach the model your schema.

ddlsql

get_tablesList all available tables.

get_schemaGet column details for a specific table.

table_name

execute_sqlRun manual SQL for verification.

sql

After setup, try these prompts with your AI agent:

→What are the top 5 products by sales volume in the inventory table?

→List all tables available in the current database.

→Show me the schema for the inventory table.

→Train the model with this DDL: CREATE TABLE users (id INT, name TEXT);

→Run a query to find all items located in bin B-12.

Prerequisites & system requirements

An MCP-compatible client (Claude Code, Cursor, Windsurf, Claude Desktop, or Codex)
Python 3.8+ with pip installed
LLM_TYPE — The LLM provider to use (ollama, lmstudio, claude, gemini, openai)

Save the docs, env vars, and workflow around Vanna AI MCP Server in Conare so Claude Code, Codex, and Cursor remember it next time.