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
| Remote MCP | Local MCP (this repo) |
|---|
| Hosting | Hosted by MotherDuck | Runs locally/self-hosted |
| Setup | Zero-setup | Requires local installation |
| Access | Read-write supported | Read-write supported |
| Local filesystem | - | Query across local and remote databases, ingest data from / export data to local filesystem |
📝 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 Code CLI commands
In-Memory DuckDB (Dev Mode):
claude mcp add --scope user duckdb --transport stdio -- uvx mcp-server-motherduck --db-path :memory: --read-write --allow-switch-databases
Local DuckDB (Read-Only):
claude mcp add --scope user duckdb --transport stdio -- uvx mcp-server-motherduck --db-path /absolute/path/to/db.duckdb
MotherDuck (Read-Write):
claude mcp add --scope user motherduck --transport stdio --env motherduck_token=YOUR_TOKEN -- uvx mcp-server-motherduck --db-path md: --read-write
Codex CLI commands
In-Memory DuckDB (Dev Mode):
codex mcp add duckdb -- uvx mcp-server-motherduck --db-path :memory: --read-write --allow-switch-databases
Local DuckDB (Read-Only):
codex mcp add duckdb -- uvx mcp-server-motherduck --db-path /absolute/path/to/db.duckdb
MotherDuck (Read-Write):
codex mcp add motherduck --env motherduck_token=YOUR_TOKEN -- uvx mcp-server-motherduck --db-path md: --read-write
Gemini CLI commands
In-Memory DuckDB (Dev Mode):
gemini mcp add -s user duckdb uvx mcp-server-motherduck --db-path :memory: --read-write --allow-switch-databases
Local DuckDB (Read-Only):
gemini mcp add -s user duckdb uvx mcp-server-motherduck --db-path /absolute/path/to/db.duckdb
MotherDuck (Read-Write):
gemini mcp add -s user -e motherduck_token=YOUR_TOKEN motherduck uvx mcp-server-motherduck --db-path md: --read-write
Kiro manual JSON config
Add the following to your Kiro MCP config file (~/.kiro/settings/mcp.json for global, or .kiro/settings/mcp.json for project-scoped). See the Kiro MCP documentation for more details.
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"]
}
}
}
MotherDuck (Read-Write):
{
"mcpServers": {
"MotherDuck (local, r/w)": {
"command": "uvx",
"args": ["mcp-server-motherduck", "--db-path", "md:", "--read-write"],
"env": {
"motherduck_token": "<YOUR_MOTHERDUCK_TOKEN>"
}
}
}
}
| Tool | Description | Required Inputs | Optional Inputs |
|---|
execute_query | Execute SQL query (DuckDB dialect) | sql | - |
list_databases | List all databases (useful for MotherDuck or multiple attached DBs) | - | - |
list_tables | List tables and views | - | database, schema |
list_columns | List columns of a table/view | table | database, schema |
switch_database_connection* | Switch to different database | path | create_if_not_exists |
*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
| Parameter | Default | Description |
|---|
--db-path | :memory: | Database path: local file (absolute), md: (MotherDuck), or s3:// URL |
--motherduck-token | motherduck_token env var | MotherDuck access token |
--read-write | False | Enable write access |
--motherduck-saas-mode | False | MotherDuck SaaS mode (restricts local access) |
--allow-switch-databases | False | Enable switch_database_connection tool |
--max-rows | 1024 | Max rows returned |
--max-chars | 50000 | Max characters returned |
--query-timeout | -1 | Query timeout in seconds (-1 = disabled) |
--init-sql | None | SQL to execute on startup |
--motherduck-connection-parameters | session_hint=mcp&
dbinstance_inactivity_ttl=0s | Additional MotherDuck connection string parameters (key=value pairs separated by &) |
--ephemeral-connections | True | Use temporary connections for read-only local files |
--transport | stdio | Transport type: stdio or http |
--stateless-http | False | For protocol compatibility only (e.g. with AWS Bedrock AgentCore Runtime). Server still maintains global state via the shared DatabaseClient. |
--port | 8000 | Port for HTTP transport |
--host | 127.0.0.1 | Host for HTTP transport |
Environment Variables
| Variable | Description |
|---|
motherduck_token or MOTHERDUCK_TOKEN | MotherDuck access token (alternative to --motherduck-token) |
HOME | Used by DuckDB for extensions and config. Override with --home-dir if not set. |
AWS_ACCESS_KEY_ID | AWS access key for S3 database connections |
AWS_SECRET_ACCESS_KEY | AWS secret key for S3 database connections |
AWS_SESSION_TOKEN | AWS session token for temporary credentials (IAM roles, SSO, EC2 instance profiles) |
AWS_DEFAULT_REGION | AWS region for S3 connections |
AWS_ENDPOINT | AWS endpoint for S3 connections |
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
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
- Run the
Release New Version GitHub Action
- Enter version in
MAJOR.MINOR.PATCH format
- 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