[Python] WebSocket connection closes when calling MCP tools in Google Cloud Run

I'm building a travel planner agent using MCP (Model Context Protocol) servers with Python and Streamlit. The application works perfectly locally but fails when deployed to Google Cloud Run with WebSocket connection issues.

Expected behavior: MCP tools should connect to external servers (Airbnb MCP, Travel Planner MCP) and return data successfully, just like they do locally.

Actual behavior: WebSocket connections close immediately in the Cloud Run environment, causing the MCP tool calls to fail.

Error Messages:
Console Error:

WebSocket connection closed unexpectedly
Connection to MCP server failed: Connection closed


Cloud Run Logs:

[ERROR] WebSocket handshake failed
[ERROR] MCP server connection timeout after 60 seconds


Environment:

1. 
   Local: Python 3.11, Windows (working)
   
   
2. 
   Cloud Run: Python 3.11, Container environment (failing)
   
   
3. 
   Libraries: streamlit, asyncio, websockets (specify exact versions)
   
   
4. 
   MCP Servers: @openbnb/mcp-server-airbnb, @gongrzhe/server-travelplanner-mcp
   

How can I configure Cloud Run to allow these MCP server connections to work properly?

Cloud Config

Service travel-planner in region us-central1

URL: https://travel-planner-809145899609.us-central1.run.app
Ingress: all
Traffic:
100% LATEST (currently travel-planner-00009-zvv)
Scaling: Auto (Min: 0)
Last updated on 2025-09-12T13:28:38.830709Z by xyz@123gmail.com:
Revision travel-planner-00009-zvv
Container None
Image: gcr.io/ai-agent-travel-planner/travel-planner
Port: 8080
Memory: 2Gi
CPU: 1000m
Secrets:
GEMINI_API_KEY GEMINI_API_KEY:latest
GOOGLE_MAPS_API_KEY GOOGLE_MAPS_API_KEY:latest
Startup Probe:
TCP every 240s
Port: 8080
Initial delay: 0s
Timeout: 240s
Failure threshold: 1
Type: Default
Service account: 809145899609-compute@developer.gserviceaccount.com
Concurrency: 80
Max instances: 20
Timeout: 300s


Example

import os
import asyncio
import MultiMCPTools

async def run_mcp_travel_planner(destination: str, num_days: int,
preferences: str, budget: int,
openai_key: str, google_maps_key: str):
try:
# Set Google Maps API key environment variable
os.environ["GOOGLE_MAPS_API_KEY"] = google_maps_key

# Initialize MCPTools with Airbnb MCP
mcp_tools = MultiMCPTools(
[
"npx -y @openbnb/mcp-server-airbnb --ignore-robots-txt",
"npx @gongrzhe/server-travelplanner-mcp",
],
env={
"GOOGLE_MAPS_API_KEY": google_maps_key,
},
timeout_seconds=60,
)

# This works locally but fails in Cloud Run
await mcp_tools.connect() # WebSocket connection closes here


except Exception as e:
print(f"MCP connection failed: {e}")
raise

# Test function
async def main():
await run_mcp_travel_planner("Paris", 3, "museums", 1000, "key", "maps_key")

if __name__ == "__main__":
asyncio.run(main())

Continue reading...