Skip to content

6-mcp-approach.md

Latest commit

 

History

History
77 lines (57 loc) · 2.18 KB

6-mcp-approach.md

File metadata and controls

77 lines (57 loc) · 2.18 KB

MCP Tools for App Lifecycle

Use MCP tools to create, deploy, and manage Databricks Apps programmatically. This mirrors the CLI workflow but can be invoked by AI agents.

Workflow

Step 1: Write App Files Locally

Create your app files in a local folder:

my_app/
├── app.py             # Main application
├── models.py          # Pydantic models
├── backend.py         # Data access layer
├── requirements.txt   # Additional dependencies
└── app.yaml           # Databricks Apps configuration

Step 2: Upload to Workspace

# MCP Tool: upload_folder
upload_folder(
    local_folder="/path/to/my_app",
    workspace_folder="/Workspace/Users/user@example.com/my_app"
)

Step 3: Create and Deploy App

# MCP Tool: create_or_update_app (creates if needed + deploys)
result = create_or_update_app(
    name="my-dashboard",
    description="Customer analytics dashboard",
    source_code_path="/Workspace/Users/user@example.com/my_app"
)
# Returns: {"name": "my-dashboard", "url": "...", "created": True, "deployment": {...}}

Step 4: Verify

# MCP Tool: get_app (with logs)
app = get_app(name="my-dashboard", include_logs=True)
# Returns: {"name": "...", "url": "...", "status": "RUNNING", "logs": "...", ...}

Step 5: Iterate

  1. Fix issues in local files
  2. Re-upload with upload_folder
  3. Re-deploy with create_or_update_app (will update existing + deploy)
  4. Check get_app(name=..., include_logs=True) for errors
  5. Repeat until app is healthy

Quick Reference: MCP Tools

Tool Description
create_or_update_app Create app if it doesn't exist, optionally deploy (pass source_code_path)
get_app Get app details by name (with include_logs=True for logs), or list all apps
delete_app Delete an app
upload_folder Upload local folder to workspace (shared tool)

Notes

  • Add resources (SQL warehouse, Lakebase, etc.) via the Databricks Apps UI after creating the app
  • MCP tools use the service principal's permissions — ensure it has access to required resources
  • For manual deployment, see 4-deployment.md