Claude Code Setup

This is the deep-dive onboarding guide for Claude Code. For a quick reference across all platforms, see the MCP Client Setup page.

Prerequisites

•Claude Code — Latest version installed. Update with claude update or reinstall from the official site.
•Node.js 20+ — Required to run MCP servers. Check with node --version.
•npm — Bundled with Node.js. Used by npx to fetch and run the Data Workers server.

Install with One Command

The fastest way to connect Data Workers to Claude Code:

claude mcp add data-workers -- npx dw-claw

This registers the full Data Workers MCP server (15 agents, 155+ tools) with Claude Code. The registration persists across sessions — you only run this once.

Manual Configuration

If you prefer manual setup, edit .mcp.json in your project root:

{ "mcpServers": { "data-workers": { "command": "npx", "args": ["dw-claw"] } } }

Save the file and restart Claude Code. The server will start automatically on the next session.

Verify Connection

Step 1: Check the server is registered:

claude mcp list

You should see data-workers with a connected status.

Step 2: Verify tools are discovered. Type in the Claude Code chat:

List all available Data Workers tools and show me the health of my data pipelines.

You should see tools organized by agent domain — incident, quality, schema, pipeline, catalog, governance, and more.

InMemory Stubs on First Run

When you first run without infrastructure credentials, the server starts with InMemory stub data. This is expected, not an error. You will see responses referencing sample datasets like analytics.orders, staging.customers, and raw.events. This confirms the server is working correctly.

To connect to real infrastructure, set environment variables (e.g., SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER) before launching Claude Code. The server auto-detects real credentials and switches from stubs to live connections.

First 5 Workflows to Try

Once connected, try these prompts in the Claude Code chat:

•Incident diagnosis: "Why did my nightly ETL pipeline fail last night? Check the logs and give me a root cause analysis."
•Data quality check: "Run a data quality assessment on the analytics.orders table and flag any anomalies."
•Catalog search: "Search the catalog for all tables related to customer revenue and show their lineage."
•Schema analysis: "Analyze the schema of staging.events and suggest improvements for query performance."
•Pipeline health: "Show me the health of all pipelines across all orchestrators and highlight any that need attention."

Individual Agent Configuration

By default, npx dw-claw starts all 15 agents. To run a single agent:

claude mcp add data-workers-incident -- npx dw-claw --agent incident

claude mcp add data-workers-quality -- npx dw-claw --agent quality

claude mcp add data-workers-schema -- npx dw-claw --agent schema

claude mcp add data-workers-pipeline -- npx dw-claw --agent pipeline

claude mcp add data-workers-context -- npx dw-claw --agent context

claude mcp add data-workers-governance -- npx dw-claw --agent governance

This is useful for resource-constrained environments or when you want to limit tool discovery to a single domain. See the Agent Reference for the complete list.

Environment Variables

Set these before launching Claude Code to connect to real infrastructure:

•SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER, SNOWFLAKE_PASSWORD — Snowflake
•GOOGLE_CLOUD_PROJECT, GOOGLE_APPLICATION_CREDENTIALS — BigQuery / Dataplex
•DATABRICKS_HOST, DATABRICKS_TOKEN — Databricks
•DBT_API_TOKEN, DBT_ACCOUNT_ID — dbt Cloud
•AIRFLOW_HOST, AIRFLOW_USER, AIRFLOW_PASSWORD — Apache Airflow
•DAGSTER_HOST, DAGSTER_TOKEN — Dagster Cloud
•PREFECT_API_KEY, PREFECT_API_URL — Prefect Cloud

The server auto-detects which credentials are present and activates the corresponding real adapters. Any service without credentials falls back to InMemory stubs.

Team Setup

To share the Data Workers configuration across your team, add a .mcp.json file to your project root and commit it to git:

{ "mcpServers": { "data-workers": { "command": "npx", "args": ["dw-claw"] } } }

Team members who clone the repo will automatically have Data Workers configured in Claude Code. Store credentials in environment variables or a .env file (add .env to .gitignore).

Removing the Server

To cleanly remove Data Workers from Claude Code:

claude mcp remove data-workers

Troubleshooting

"Agent not found" when running `claude mcp add` — Update Claude Code to the latest version. Older versions may not support the Data Workers MCP servers.

`claude mcp list` shows disconnected — The npx process may have exited. Run npx dw-claw manually in a terminal to see the full error output. Common causes: missing Node.js, network issues, or a stale npx cache (npx --yes dw-claw to force fresh download).

No tools discovered — Restart Claude Code after adding the server. Run claude mcp list to verify the server appears and is connected.

Tools appear but return credential errors — Set the required environment variables for your infrastructure before launching Claude Code. Without credentials, the server uses InMemory stubs.

Multiple MCP servers — Claude Code supports multiple MCP servers simultaneously. Data Workers runs alongside other MCP servers without conflict.

Still stuck? Join our Discord at discord.com/invite/b8DR5J53 or open an issue on GitHub.

MCP Client Setup Cursor Setup