subsetsio/subsets-mcp-server

Subsets MCP Server

Model Context Protocol server for querying statistical datasets with AI assistants.

Quick Start

Using Claude Desktop

Add to your Claude Desktop config:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "subsets": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/subsetsio/subsets-mcp-server.git",
        "mcp-server",
        "--api-key",
        "YOUR_API_KEY"
      ]
    }
  }
}

Get your API key at subsets.io/settings

Requires uv installed.

Using Other MCP Clients

uvx --from git+https://github.com/subsetsio/subsets-mcp-server.git mcp-server --api-key YOUR_API_KEY

Available MCP Tools

list_datasets

Search and browse available datasets with semantic search.

Parameters:

• q (string): Search query
• limit (integer): Max results (default: 10)
• min_score (float): Min relevance score threshold (0.0-2.0)

Example:

list_datasets(q="unemployment europe", limit=5)

get_dataset_details

Get detailed information about a specific dataset including schema, statistics, and preview.

Parameters:

• dataset_id (string): Dataset identifier

Returns: Full metadata, column descriptions, row counts, data preview, and query usage stats

Example:

get_dataset_details("eurostat_unemployment_2024")

execute_sql_query

Run SQL queries on datasets using DuckDB.

Parameters:

• query (string): SQL SELECT statement

Example:

execute_sql_query("SELECT * FROM eurostat_unemployment_2024 LIMIT 10")

CLI Tools

The package also includes CLI commands for managing local datasets:

# Install globally
npm install -g @subsetsio/mcp-server

# Add datasets to local collection
subsets add eurostat_unemployment_2024

# Download datasets
subsets sync

# List local datasets
subsets list

# View status
subsets status

Local Development

# Clone repository
git clone https://github.com/subsetsio/subsets-mcp-server
cd subsets-mcp-server

# Install dependencies
uv sync

# Run MCP server
uv run python src/server.py --api-key YOUR_API_KEY

# Or run CLI
uv run python src/cli.py --help

Documentation

Full documentation at subsets.io/docs

License

MIT