Point the AI at a local SQLite file and ask questions: tables, columns, trends, duplicates. No GUI tool needed.
Use case: dig into a .db/.sqlite export (local analytics, an app's database, a client file) without installing a GUI tool.
{
"mcpServers": {
"sqlite": {
"command": "uvx",
"args": ["--with", "mcp<2", "mcp-server-sqlite", "--db-path", "~/data/app.db"]
}
}
}
This server is Python:
uvx(ships with uv,brew install uv) runs it without installing. The--with mcp<2pin is required: this archived reference server breaks with the MCP 2.x SDK (AttributeError: 'Server' object has no attribute 'list_resources'). Node alternative, no Python:npx -y mcp-sqlite ~/data/app.db.⚠️ The path must point at an existing database: SQLite silently creates an empty file when the path is new — the model will then answer "no tables". Check with
lsbefore configuring.
Exposed tools: read_query, write_query, list_tables,
describe_table, create_table…
List the tables in this database, describe the columns of the 3 largest,
and tell me what this database seems to store.
What are the 10 most frequent events in the events table over the last
30 days? Add the trend versus the previous month.
Find the duplicates in the contacts table (same email), show them to me,
then propose the DELETE query — do not run it without my approval.
tools=sqlite.read_query = read-only guardrail.