DuckDB / MotherDuck Local MCP Server

SQL analytics and data engineering for AI Assistants and IDEs.

Connect AI assistants to your data using DuckDB's powerful analytical SQL engine. Supports connecting to local DuckDB files, in-memory databases, S3-hosted databases, and MotherDuck. Allows executing SQL read- and write-queries, browsing database catalogs, and switching between different database connections on-the-fly.

Looking for a fully-managed remote MCP server for MotherDuck? → Go to the MotherDuck Remote MCP docs

Remote vs Local MCP

📝 Migrating from v0.x?

• Read-only by default: The server now runs in read-only mode by default. Add --read-write to enable write access. See Securing for Production.
• Default database changed: --db-path default changed from md: to :memory:. Add --db-path md: explicitly for MotherDuck.
• MotherDuck read-only requires read-scaling token: MotherDuck connections in read-only mode require a read-scaling token. Regular tokens require --read-write.

Quick Start

Prerequisites: Install uv via pip install uv or brew install uv

Connecting to In-Memory DuckDB (Dev Mode)

{
  "mcpServers": {
    "DuckDB (in-memory, r/w)": {
      "command": "uvx",
      "args": ["mcp-server-motherduck", "--db-path", ":memory:", "--read-write", "--allow-switch-databases"]
    }
  }
}

Full flexibility with no guardrails — read-write access and the ability to switch to any database (local files, S3, or MotherDuck) at runtime.

Connecting to a Local DuckDB File in Read-Only Mode

{
  "mcpServers": {
    "DuckDB (read-only)": {
      "command": "uvx",
      "args": ["mcp-server-motherduck", "--db-path", "/absolute/path/to/your.duckdb"]
    }
  }
}

Connects to a specific DuckDB file in read-only mode. Won't hold on to the file lock, so convenient to use alongside a write connection to the same DuckDB file. You can also connect to remote DuckDB files on S3 using s3://bucket/path.duckdb — see Environment Variables for S3 authentication. If you're considering third-party access to the MCP, see Securing for Production.

Connecting to MotherDuck in Read-Write Mode

{
  "mcpServers": {
    "MotherDuck (local, r/w)": {
      "command": "uvx",
      "args": ["mcp-server-motherduck", "--db-path", "md:", "--read-write"],
      "env": {
        "motherduck_token": "<YOUR_MOTHERDUCK_TOKEN>"
      }
    }
  }
}

See Command Line Parameters for more options, Securing for Production for deployment guidance, and Troubleshooting if you encounter issues.

Client Setup

Client	Config Location	One-Click Install
Claude Desktop	Settings → Developer → Edit Config	.mcpb (MCP Bundle)
Claude Code	Use CLI commands below	-
Codex CLI	Use CLI commands below or ~/.codex/config.toml	-
Gemini CLI	Use CLI commands below or ~/.gemini/settings.json	-
Cursor	Settings → MCP → Add new global MCP server
VS Code	Ctrl+Shift+P → "Preferences: Open User Settings (JSON)"
Kiro	~/.kiro/settings/mcp.json (global) or .kiro/settings/mcp.json (project)

Any MCP-compatible client can use this server. Add the JSON configuration from Quick Start to your client's MCP config file. Consult your client's documentation for the config file location.

claude mcp add --scope user duckdb --transport stdio -- uvx mcp-server-motherduck --db-path :memory: --read-write --allow-switch-databases

claude mcp add --scope user duckdb --transport stdio -- uvx mcp-server-motherduck --db-path /absolute/path/to/db.duckdb

claude mcp add --scope user motherduck --transport stdio --env motherduck_token=YOUR_TOKEN -- uvx mcp-server-motherduck --db-path md: --read-write

codex mcp add duckdb -- uvx mcp-server-motherduck --db-path :memory: --read-write --allow-switch-databases

codex mcp add duckdb -- uvx mcp-server-motherduck --db-path /absolute/path/to/db.duckdb

codex mcp add motherduck --env motherduck_token=YOUR_TOKEN -- uvx mcp-server-motherduck --db-path md: --read-write

gemini mcp add -s user duckdb uvx mcp-server-motherduck --db-path :memory: --read-write --allow-switch-databases

gemini mcp add -s user duckdb uvx mcp-server-motherduck --db-path /absolute/path/to/db.duckdb

gemini mcp add -s user -e motherduck_token=YOUR_TOKEN motherduck uvx mcp-server-motherduck --db-path md: --read-write

{
  "mcpServers": {
    "DuckDB (in-memory, r/w)": {
      "command": "uvx",
      "args": ["mcp-server-motherduck", "--db-path", ":memory:", "--read-write", "--allow-switch-databases"]
    }
  }
}

{
  "mcpServers": {
    "MotherDuck (local, r/w)": {
      "command": "uvx",
      "args": ["mcp-server-motherduck", "--db-path", "md:", "--read-write"],
      "env": {
        "motherduck_token": "<YOUR_MOTHERDUCK_TOKEN>"
      }
    }
  }
}

Tools

*Requires --allow-switch-databases flag

All tools return JSON. Results are limited to 1024 rows / 50,000 chars by default (configurable via --max-rows, --max-chars).

Securing for Production

When giving third parties access to a self-hosted MCP server, read-only mode alone is not sufficient — it still allows access to the local filesystem, changing DuckDB settings, and other potentially sensitive operations.

For production deployments with third-party access, we recommend MotherDuck Remote MCP — zero-setup, read-write capable, and hosted by MotherDuck.

Self-hosting MotherDuck MCP: Fork this repo and customize as needed. Use a service account with read-scaling tokens and enable SaaS mode to restrict local file access.

Self-hosting DuckDB MCP: Use --init-sql to apply security settings. See the Securing DuckDB guide for available options.

Docker

Build and run the server with Streamable HTTP on port 8000 (defaults to an in-memory DuckDB):

docker build -t mcp-server-motherduck .
docker run --rm -p 8000:8000 mcp-server-motherduck

Connect to MotherDuck by passing a token and overriding the command:

docker run --rm -p 8000:8000 \
  -e motherduck_token="$MOTHERDUCK_TOKEN" \
  mcp-server-motherduck --transport http --db-path md:

The MCP endpoint is available at http://localhost:8000/mcp. CLI flags and env vars from below still apply.

Command Line Parameters

Environment Variables

Troubleshooting

• spawn uvx ENOENT: Specify full path to uvx (run which uvx to find it)
• File locked: Make sure --ephemeral-connections is turned on (default: true) and that you're not connected in read-write mode

Resources

• MotherDuck MCP Documentation
• Close the Loop: Faster Data Pipelines with MCP, DuckDB & AI (Blog)
• Faster Data Pipelines with MCP and DuckDB (YouTube)

Development

To run from source:

{
  "mcpServers": {
    "Local DuckDB (Dev)": {
      "command": "uv",
      "args": ["--directory", "/path/to/mcp-server-motherduck", "run", "mcp-server-motherduck", "--db-path", "md:"],
      "env": {
        "motherduck_token": "<YOUR_MOTHERDUCK_TOKEN>"
      }
    }
  }
}

Release Process

1. Run the Release New Version GitHub Action
2. Enter version in MAJOR.MINOR.PATCH format
3. The workflow bumps version, publishes to PyPI/MCP registry, and creates the GitHub release with MCPB package

License

MIT License - see LICENSE file.

mcp-name: io.github.motherduckdb/mcp-server-motherduck