DuckDB MCP Server

Content

DuckDB MCP Server is a Python MCP server that connects Claude to a local DuckDB database file. It exposes one tool, query, which executes SQL against the configured database and returns results as text.

The server is intentionally simple: there are no custom MCP resources or prompts, and database interaction is handled through SQL. That makes it useful for local analytics, schema inspection, and data exploration, but it also means the configured read/write mode matters. The README documents a --readonly flag that opens DuckDB with native read-only protection and prevents write operations.

Source Review

• https://github.com/ktanaka101/mcp-server-duckdb
• https://github.com/ktanaka101/mcp-server-duckdb/blob/main/README.md
• https://pypi.org/project/mcp-server-duckdb/
• https://github.com/ktanaka101/mcp-server-duckdb/blob/main/pyproject.toml
• https://github.com/ktanaka101/mcp-server-duckdb/blob/main/src/mcp_server_duckdb/server.py
• https://github.com/ktanaka101/mcp-server-duckdb/blob/main/LICENSE

These sources were reviewed on 2026-06-06. Prefer the live repository, README, PyPI page, package metadata, server implementation, and license for current command names, package version, database flags, and tool behavior.

Features

• Stdio MCP server packaged as mcp-server-duckdb.
• Single query tool for executing DuckDB SQL.
• Configurable --db-path for the target DuckDB database file.
• Optional --readonly mode backed by DuckDB's native read-only connection.
• Automatic database file and parent-directory creation when not in read-only mode.
• Optional --keep-connection mode for reusing a single connection across requests.
• Query results returned as text to the MCP client.
• No custom MCP resources or prompts.

Installation

For MCP clients that launch stdio servers with uvx:

{
  "mcpServers": {
    "duckdb": {
      "command": "uvx",
      "args": [
        "mcp-server-duckdb",
        "--db-path",
        "./data.duckdb",
        "--readonly"
      ]
    }
  }
}

Remove --readonly only when Claude is allowed to create or mutate the database file. Replace ./data.duckdb with the approved database path for the project.

Use Cases

• Ask Claude to inspect schema and sample analytical tables from a local DuckDB file.
• Run reviewed SELECT queries over local analytics data, evaluation outputs, CSV imports, or Parquet-derived tables.
• Let Claude draft SQL while a human approves the query before execution.
• Use read-only mode for safer data exploration in model-assisted workflows.
• Use a writable disposable database when prototyping transformations or temporary tables.

Safety and Privacy

DuckDB MCP gives a model a SQL execution surface. Use --readonly by default for existing data, and only switch to writable mode for disposable or backed-up databases where mutation is intended. Review generated SQL before execution, especially statements that create, insert, update, delete, attach files, install extensions, or reference local paths.

Keep database files, source data directories, and query outputs scoped to what the model is allowed to see. Even read-only queries can reveal sensitive rows, schemas, filenames, derived metrics, or internal business logic through results and errors.

Duplicate Check

The catalog already has a DuckDB tools entry for the embedded analytical database engine itself. This entry covers the separate ktanaka101/mcp-server-duckdb MCP wrapper that exposes DuckDB through Model Context Protocol. No mcp-server-duckdb, ktanaka101/mcp-server-duckdb, or dedicated DuckDB MCP entry was found in content/mcp, content/guides, content/agents, or content/skills.