databricks-transform-mcp

A Databricks App that hosts the Coalesce Transform MCP server over HTTP so Databricks Genie can connect to it.

How it works

Genie / AI Agent
      │
      │  HTTPS  (Databricks App proxy handles auth)
      ▼
Databricks App  (uvicorn + server.py)
      │
      │  Python MCP SDK bridges HTTP ↔ stdio
      ▼
coalesce-transform-mcp  (npm package, 100+ tools)
      │
      ▼
Coalesce API  +  Snowflake

server.py uses the Python MCP SDK to spin up coalesce-transform-mcp as a subprocess, then exposes its tools over Streamable HTTP via a Starlette/uvicorn ASGI app.

Related MCP server: Snowflake Cortex AI MCP Server

Prerequisites

• Databricks workspace with Apps enabled

• Coalesce access token (Coalesce Deploy → Settings → API Tokens)

• Node.js ≥ 22 on the App cluster (needed to run npx coalesce-transform-mcp)

• Python ≥ 3.11

1. Store the Coalesce token in Databricks Secrets

# Create a secret scope (skip if it already exists)
databricks secrets create-scope coalesce-mcp

# Store your Coalesce API token
databricks secrets put-secret coalesce-mcp coalesce-access-token
# paste your token when prompted

2. Link the secret to the app via the UI

Databricks Apps injects secrets as env vars only after they're linked to the app through the UI:

1. In Databricks, go to Compute → Apps and open (or create) the app

2. Click Edit app → App resources → + Add resource → Secret

3. Set:

   • Scope: coalesce-mcp

   • Key: coalesce-access-token

   • Resource key: coalesce-access-token ← must match valueFrom in app.yaml

   • Permission: Can read

4. Save and redeploy

This is what wires the COALESCE_ACCESS_TOKEN env var in app.yaml to the actual secret value.

3. Deploy

Push this repo to a Databricks Git Folder, then deploy:

databricks apps deploy coalesce-transform-mcp \
  --source-code-path /Workspace/Repos/<your-user>/databricks-transform-mcp

Or via the Apps UI:

1. Compute → Apps → Create App → Custom

2. Point to this Git Folder

3. Click Deploy

On first deploy, pip install -r requirements.txt runs automatically, then uvicorn starts.

3. Connect Genie

1. In Genie, go to Settings → MCP Servers → Add

2. Set the endpoint to the Streamable HTTP path:

   https://<workspace-host>/apps/coalesce-transform-mcp/mcp

3. Auth is handled automatically by the Databricks App proxy

Environment Variables

Local Testing

pip install -r requirements.txt

export COALESCE_ACCESS_TOKEN=your-token
export COALESCE_BASE_URL=https://app.australia-southeast1.gcp.coalescesoftware.io

uvicorn server:app --host 0.0.0.0 --port 8000

Test the MCP endpoint:

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'