MCP SQL Query — SQLite Database Operations for AI Agents

An MCP server that gives AI agents SQLite database access: execute SQL, inspect schemas, CRUD operations, export to JSON/CSV — 13 tools, zero dependencies, pure Python stdlib.

Features

• 13 MCP Tools: execute_sql, execute_batch, list_tables, table_schema, create_table, insert_row, insert_batch, select_rows, update_rows, delete_rows, export_json, export_csv, database_info

• Zero dependencies — pure Python stdlib (sqlite3, json, re, os)

• SQL injection prevention — table name validation with regex

• Batch transactions — execute multiple statements atomically with rollback

• Data export — JSON array or CSV string output

• In-memory or file-based — use :memory: for ephemeral DBs or file paths for persistence

• STDIO JSON-RPC mode — drop-in for Claude Desktop, Hermes, or any MCP client

Quick Start

python -m src.server --stdio    # STDIO mode
python -m src.server --manifest # Print manifest

Use as a library

from src.server import MCPSQLQueryServer
import json

server = MCPSQLQueryServer()

# Create a table
server.handle_tool_call("create_table", {
    "table_name": "users",
    "columns": [
        {"name": "id", "type": "INTEGER", "primary_key": True},
        {"name": "name", "type": "TEXT", "not_null": True},
        {"name": "email", "type": "TEXT", "unique": True}
    ]
})

# Insert rows
server.handle_tool_call("insert_row", {
    "table_name": "users",
    "data": {"name": "Alice", "email": "alice@example.com"}
})

# Query
result = server.handle_tool_call("select_rows", {
    "table_name": "users",
    "where": "name = 'Alice'"
})

# Export
server.handle_tool_call("export_csv", {"table_name": "users"})

MCP Tool Reference

Tests

python -m pytest tests/ -v  # 33 tests, all passing

License

MIT

Author

AMEOBIUS — github.com/AMEOBIUS