Sqlite

sqlite

An MCP server that exposes a small, agent-friendly toolset for working with SQLite databases. Writes are disabled by default for safety and must be explicitly enabled.

It ships a runnable MCP server instead of a SKILL.md.

What it does

The server registers five tools with the model:

Tool	Writes?	Description
query	no	Execute a SELECT query and return rows as a list of dicts (with row_count). Non-SELECT statements are rejected.
list_tables	no	List all tables in the database.
describe_table	no	Return a table's schema: columns (name, type, nullable, default, primary_key), the CREATE TABLE SQL, and indexes.
execute	yes	Execute an INSERT, UPDATE, or DELETE and return the affected row count. Requires writes enabled.
create_table	yes	Create a new table from a list of column definitions. Requires writes enabled.

Once the plugin is enabled, the tools are namespaced by the plugin loader as mcp__plugin_sqlite_sqlite__<tool> (e.g. mcp__plugin_sqlite_sqlite__query). Verify the exact id by inspecting the available tools after installing the plugin.

Read-only by default

execute and create_table return an error unless write operations are enabled. To enable writes, set the environment variable when the server is launched:

• Environment variable: MCP_SQLITE_ALLOW_WRITES=true
• CLI flag (when run standalone): --allow-writes

The bundled .mcp.json intentionally launches the server without --allow-writes, keeping the safe read-only default. To allow writes, add an env block to the server entry, for example:

{
  "mcpServers": {
    "sqlite": {
      "command": "uv",
      "args": ["run", "--script", "${CLAUDE_PLUGIN_ROOT}/scripts/mcp_sqlite_server.py"],
      "env": { "MCP_SQLITE_ALLOW_WRITES": "true" }
    }
  }
}

Note: even with writes disabled, opening a database file that does not yet exist is refused (SQLite would otherwise create an empty file).

Database path (db_path)

Every tool accepts an optional db_path argument. When omitted, the server's default database (configured at launch via --db-path) is used. Supported forms:

• Relative: data.db, ./databases/app.db
• Absolute: /var/lib/myapp/data.db
• In-memory: :memory: — a temporary database whose state persists for the lifetime of the server process
• Home-relative: ~/myapp/data.db — expands to the user's home directory

To configure a default database, add --db-path /path/to/db.sqlite to the args array in .mcp.json.

Prerequisites

• uv — the server is launched with uv run --script. The script is a PEP 723 inline-dependency script (requires-python >=3.10, mcp>=0.1.0); uv resolves and caches the dependency automatically on first run. No manual pip install is needed.
• No database engine to install — sqlite3 is part of the Python standard library, so there is no external binary prerequisite.

How it is wired

The plugin declares the server in .mcp.json:

{
  "mcpServers": {
    "sqlite": {
      "command": "uv",
      "args": ["run", "--script", "${CLAUDE_PLUGIN_ROOT}/scripts/mcp_sqlite_server.py"]
    }
  }
}

${CLAUDE_PLUGIN_ROOT} is expanded by Claude Code to the installed plugin directory, so the server runs regardless of where the plugin is checked out.

Tests

cd plugins/sqlite
uv run --isolated --extra dev pytest tests/ -v