SQLite MCP Server: Install Guide and Examples (2026)

Name: SQLite MCP Server
Author: Anthropic

How to install the SQLite MCP server

Add this to your Claude Desktop MCP configuration:

{
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": [
        "mcp-server-sqlite",
        "--db-path",
        "/path/to/db.sqlite"
      ]
    }
  }
}

Add this to your Claude Code MCP configuration:

uvx mcp-server-sqlite --db-path /path/to/db.sqlite

Add this to your Cursor MCP configuration:

{
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": [
        "mcp-server-sqlite",
        "--db-path",
        "/path/to/db.sqlite"
      ]
    }
  }
}

Add this to your Windsurf MCP configuration:

{
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": [
        "mcp-server-sqlite",
        "--db-path",
        "/path/to/db.sqlite"
      ]
    }
  }
}

Add this to your Cline MCP configuration:

{
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": [
        "mcp-server-sqlite",
        "--db-path",
        "/path/to/db.sqlite"
      ]
    }
  }
}

Add this to your VS Code (Continue) MCP configuration:

{
  "mcpServers": {
    "sqlite": {
      "command": "uvx",
      "args": [
        "mcp-server-sqlite",
        "--db-path",
        "/path/to/db.sqlite"
      ]
    }
  }
}

Built by ContextBoltThis directory is built by ContextBolt: MCP-native memory and SEO tools that run alongside SQLite in the same client.

Explore ContextBolt

The SQLite MCP server is Anthropic’s official way to let Claude query a local SQLite file. You give it a path to a .db or .sqlite file and Claude can inspect the schema, run queries and write results back. It is one of the smallest useful servers in the ecosystem and a good demonstration of why MCP beats copy-paste for structured data.

What it actually does

The server exposes the database as a set of tools: list the tables, describe a table, run a read query, run a write. Because Claude can read the schema first, it writes queries against real column names instead of inventing them, which is the failure mode when you paste a fragment of a table into a chat and hope for the best.

Practical patterns:

‘What tables are in this database and how do they relate?’
‘How many rows were added each month last year?’
‘Find the customers with no orders in the last 90 days.’
‘Add an index that would speed up that query.‘

Why use it

SQLite files turn up everywhere: analytics exports, app state, scraped datasets, the local cache half your tools keep. They are awkward to explore without a client and tedious to summarise by hand. A server that can read the schema and answer in plain language removes the step where you go and remember the syntax.

Gotchas

Write access is real and unguarded. The server will happily run a statement that drops a table if the conversation leads there, so work against a copy of anything you care about. Large result sets also land in the context window, so ask for aggregates rather than “show me everything” on a table with a million rows. And it is genuinely SQLite-only; pointing it at a Postgres connection string will not work.

Built by ContextBolt

This directory is built by ContextBolt

We build MCP-native tools that give your AI the context it cannot reach on its own. Bookmarks turns your saved posts into agent-queryable memory. SEO puts live keyword and ranking data inside Claude. Both run alongside SQLite in the same client.

Explore the products →