Skip to main content

Connect an MCP server

This recipe connects to an MCP server for AI applications using GitHub as an example and Unity Catalog HTTP connections for secure and governed access.

Code snippets​

OAuth User to Machine Per User (On-behalf-of-user)​

app.py
import streamlit as st
from databricks.sdk import WorkspaceClient
from databricks.sdk.service.serving import ExternalFunctionRequestHttpMethod
import json


token = st.context.headers.get("x-forwarded-access-token")
w = WorkspaceClient(token=token, auth_type="pat")


def init_mcp_session(w: WorkspaceClient, connection_name: str):
    init_payload = {
        "jsonrpc": "2.0",
        "id": "init-1",
        "method": "initialize",
        "params": {}
    }
    response = w.serving_endpoints.http_request(
        conn=connection_name,
        method=ExternalFunctionRequestHttpMethod.POST,
        path="/",
        json=init_payload,
    )
    return response.headers.get("mcp-session-id")


connection_name = "github_u2m_connection"
http_method = ExternalFunctionRequestHttpMethod.POST
path = "/"
headers = {"Content-Type": "application/json"}
payload = {"jsonrpc": "2.0", "id": "list-1", "method": "tools/list"}

if st.button("Run"):
    session_id = init_mcp_session(w, connection_name)
    headers["Mcp-Session-Id"] = session_id

    response = w.serving_endpoints.http_request(
        conn=connection_name,
        method=http_method,
        path=path,
        headers=headers,
        json=payload,
    )
    st.json(response.json())

Bearer token​

app.py
import streamlit as st
from databricks.sdk import WorkspaceClient
from databricks.sdk.service.serving import ExternalFunctionRequestHttpMethod


w = WorkspaceClient()

response = w.serving_endpoints.http_request(
    conn="github_u2m_connection",
    method=ExternalFunctionRequestHttpMethod.GET,
    path="/traffic/views",
    headers={"Accept": "application/vnd.github+json"},
    json={
        "jsonrpc": "2.0",
        "id": "init-1",
        "method": "initialize",
        "params": {}
    },
)

st.json(response.json())

Resources​

Permissions​

Your app service principal needs the following permissions:

  • USE CONNECTION permission on the HTTP Connection

When using OAuth User to Machine Per User (On-behalf-of-user), you need to configure User authorization by adding the Unity Catalog connection or other scopes.

Dependencies​

requirements.txt
databricks-sdk
streamlit
mcp[cli]