Configuration
AI Dev Kit’s MCP server uses standard Databricks authentication. Choose the method that fits your workflow.
Authentication Methods
Section titled “Authentication Methods”Set environment variables directly. Simplest option for individual use.
{ "mcpServers": { "ai-dev-kit": { "command": "uvx", "args": ["databricks-ai-dev-kit"], "env": { "DATABRICKS_HOST": "https://your-workspace.cloud.databricks.com", "DATABRICKS_TOKEN": "dapi..." } } }}Generate a token in your workspace under Settings > Developer > Access tokens.
Use an existing CLI profile. Best when you already have databricks auth login configured.
# Authenticate (one-time setup)databricks auth login --host https://your-workspace.cloud.databricks.com{ "mcpServers": { "ai-dev-kit": { "command": "uvx", "args": ["databricks-ai-dev-kit"], "env": { "DATABRICKS_CONFIG_PROFILE": "your-profile-name" } } }}The profile name matches what you set during databricks auth login. Use DEFAULT if you did not specify one.
Use a service principal for CI/CD or shared environments.
{ "mcpServers": { "ai-dev-kit": { "command": "uvx", "args": ["databricks-ai-dev-kit"], "env": { "DATABRICKS_HOST": "https://your-workspace.cloud.databricks.com", "DATABRICKS_CLIENT_ID": "your-client-id", "DATABRICKS_CLIENT_SECRET": "your-client-secret" } } }}Create a service principal in Settings > Identity and access > Service principals, then generate an OAuth secret.
MCP Server Options
Section titled “MCP Server Options”The MCP server accepts additional environment variables to customize behavior:
| Variable | Description | Default |
|---|---|---|
DATABRICKS_HOST | Workspace URL | Required (unless using profile) |
DATABRICKS_TOKEN | Personal access token | — |
DATABRICKS_CONFIG_PROFILE | CLI profile name | — |
DATABRICKS_CLIENT_ID | OAuth client ID (M2M) | — |
DATABRICKS_CLIENT_SECRET | OAuth client secret (M2M) | — |
DATABRICKS_CLUSTER_ID | Default cluster for command execution | Auto-selected |
DATABRICKS_WAREHOUSE_ID | Default SQL warehouse | Auto-selected |
When DATABRICKS_CLUSTER_ID or DATABRICKS_WAREHOUSE_ID are not set, the MCP server automatically selects an available compute resource.
Multiple Workspaces
Section titled “Multiple Workspaces”To connect to more than one workspace, add multiple MCP server entries with different names:
{ "mcpServers": { "ai-dev-kit-dev": { "command": "uvx", "args": ["databricks-ai-dev-kit"], "env": { "DATABRICKS_CONFIG_PROFILE": "dev" } }, "ai-dev-kit-prod": { "command": "uvx", "args": ["databricks-ai-dev-kit"], "env": { "DATABRICKS_CONFIG_PROFILE": "prod" } } }}Troubleshooting
Section titled “Troubleshooting”| Issue | Solution |
|---|---|
DATABRICKS_PROFILE not recognized | Use DATABRICKS_CONFIG_PROFILE — the SDK requires the full variable name |
| Token expired | Re-generate under Settings > Developer > Access tokens |
| CLI profile not found | Run databricks auth login to create or refresh the profile |
| Permission denied on catalog/table | Verify Unity Catalog grants for your user or service principal |