Issue with integrating of sample mcp-server-hello-world MCP server with Databricks Assistant

[norbitek]

I tried to add mcp-server-hello-world as new tool to Databricks Assistant
I successfully deployed this server as App
It works smoothly in the playground

But when I try to add this Custom MCP server to the Databricks Assistant I got error:

I added some logging info in the server app.py to see more details
Additionally I added CORS Configuration

combined_app.add_middleware( CORSMiddleware, allow_origins=["*"], # Allow all origins allow_credentials=True, allow_methods=["*"], # Handles OPTIONS, POST, GET allow_headers=["*"], )

Based on assistant log observation :

"It's a protocol incompatibility:

Playground client: Sends mcp-session-id header → fastmcp recognizes session → works
Assistant client: Doesn't send mcp-session-id header → fastmcp creates new session → fails with 400

Playground (working):

First POST → Creates session bd0d742b... → 200 OK
Second POST → "Session already exists, handling request directly" → 202 Accepted
Third POST → Works fine
Assistant (failing):

First POST → Creates session 47d80107... → 200 OK
Second POST → Creates NEW session 401be9da... → 400 Bad Request ❌
"

Did somebody try to add this server as Databricks assistance tool?

[wpflueger]

After fixing the CORS issue, I was eventually able to resolve the root cause by adding the following param to the mcp_server:

mcp_app = mcp_server.http_app(stateless_http=True)

Two fixes were needed:

1. CORS Middleware — Add CORS middleware to your combined app (as described above).

2. Stateless Mode — The Databricks Assistant doesn't send a mcp-session-id header with each request (unlike the Playground), so fastmcp treats every request as a new session, causing a 400 Bad Request on the second call. Initializing the MCP app in stateless mode fixes this.

The default template uses a stateful configuration, but the Workspace Assistant requires stateless mode to function correctly.

[norbitek]

Thanks for the fast response
I added this configuration and now it works

Additionally I sent a feedback to Databricks from my workspace with suggested documentation updates and update of the the available GitHub template to include both configurations

I'm not sure if my feedback will be taken into consideration or not. Will see