Unit 2: Testing Your MCP Servers

Hands-On Testing (30 minutes)

Learning Objectives

Configure the hotel app for authenticated chat
Test MCP tools as both Guest and Manager personas
Use debug mode to observe tool call activity
Troubleshoot using Workato execution logs
(Optional) Test via Claude Desktop or ChatGPT Desktop

Part 1: Enable Chat in the Hotel App (10 min)

Now that your MCP servers are created, you’ll configure the hotel app to use them with authenticated chat.

1.1 Switch to AWS Cognito Authentication

Update your app/.env file with the following changes:

# TODO: Add specific Cognito configuration variables
# Your facilitator will provide these values

1.2 Create Your Cognito Account

Restart the hotel app (if running):

app/scripts/dev-tools/server.sh restart

Open http://localhost:3000
The app will guide you through creating an AWS Cognito account
This provisions LLM access via Amazon Bedrock

1.3 Add Facilitator-Provided Keys

Your facilitator will provide temporary keys for the workshop. Add them to your app/.env:

# TODO: Add Bedrock/Cognito keys provided by facilitator

After configuration is complete:

Restart the app
The login page will now show static login options:
- Guest accounts (use your +guest email alias credentials)
- Manager accounts (use your +manager email alias credentials)

CHECKPOINT: Login page shows Guest and Manager login options

Part 2: Test as a Guest (10 min)

2.1 Log In as Guest

Select a Guest login option
Enter your credentials
Navigate to Chat in the left sidebar
Wait for the agent greeting message (confirms Bedrock is connected)

2.2 Enable Debug Mode

Look for the Debug toggle in the chat interface
Enable it to open the tool call activity panel on the right
This shows real-time MCP tool invocations

2.3 Test Guest Scenarios

Try these prompts and observe the tool calls in the debug panel:

Room Information:

What room am I staying in?

Expected tool: search_rooms_on_behalf_of_guest

Service Request:

I need extra towels in my room

Expected tools: submit_guest_service_request
Watch: How does the agent gather missing info (room number, priority)?

Check Service Status:

What's the status of my service requests?

Expected tool: search_cases_on_behalf_of_guest

Booking Management:

I'd like to extend my stay by one night

Expected tool: manage_booking_orchestrator

2.4 Observe in Workato Logs

Open Workato → Tools → Logs
Find your recent tool executions
Click on a log entry to see:
- Input parameters received
- Each step’s execution
- Response returned to the agent

CHECKPOINT: Successfully tested guest scenarios with debug panel showing tool calls

Part 3: Test as a Manager (5 min)

3.1 Switch to Manager Persona

Log out of the guest account
Log in with a Manager account
Navigate to Chat

3.2 Test Manager-Only Scenarios

These tools are only available to managers:

View All Rooms:

Show me all vacant rooms

Expected tool: search_rooms_on_behalf_of_staff
Note: Returns ALL rooms, not just guest’s bookings

Maintenance Request:

Room 205 has a leaky faucet, please file a maintenance request

Expected tool: submit_maintenance_request

Process Refund:

Process a refund for the failed checkout on booking BK-12345

Expected tool: compensate_checkout_failure

Manage Cases:

Show me all open service cases

Expected tool: search_cases_on_behalf_of_staff

3.3 Compare Guest vs Manager Access

Notice the difference:

Guest search_rooms returns only their booked rooms
Manager search_rooms returns all hotel rooms with guest details
Manager has access to maintenance, case management, and refund tools

CHECKPOINT: Confirmed manager has elevated access and additional tools

Part 4: Trigger Error Conditions (5 min)

Test how the system handles errors gracefully:

4.1 Resource Not Found (404)

Check in guest with email nonexistent@example.com

Expected: Clear error message, no crash
Check Workato logs for the 404 response

4.2 Invalid State (409)

Check in a guest who is already checked in

Expected: Conflict error explaining the issue

4.3 Missing Required Info

I need housekeeping

Expected: Agent asks for room number, not a tool error
Observe: LLM gathers required fields before calling tool

Alternative: Test via Desktop Clients

If the hotel app has issues, test your MCP servers directly:

Claude Desktop

Copy the Developer MCP Token from your MCP server’s Settings tab in Workato
Add to claude_desktop_config.json:

{
  "mcpServers": {
    "dewy-guest-mcp": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://your_url?wkt_token=your_token"
      ]
    }
  }
}

Restart Claude Desktop
Test with: “What tools do you have available?”

ChatGPT Desktop

Open Settings → MCP Servers
Add your MCP server URL and token
Test the same scenarios

Workato API Platform (Direct)

Go to Tools → API Platform → API Collections
Click on your collection
Select an endpoint
Use the Test tab to send requests directly

Troubleshooting

Issue	Solution
No agent greeting in chat	Bedrock provisioning incomplete; check env vars
Login page unchanged	Restart app after env changes
Tool calls not showing	Enable Debug mode toggle
“Unauthorized” errors	Verify MCP server tokens in .env
Tool returns error	Check Workato Tools → Logs for details

What You Accomplished

Configured authenticated chat with AWS Cognito + Bedrock
Tested MCP tools as both Guest and Manager personas
Used debug mode to observe real-time tool activity
Explored error handling and LLM coaching behavior
(Optional) Validated MCP servers via desktop clients

Facilitator Notes

Before this unit:

Ensure all attendees have MCP servers created from Unit 1
Have temporary Cognito/Bedrock keys ready to distribute
Test the full flow yourself before the session

Common issues:

Attendees forgetting to restart app after env changes
Bedrock provisioning delays (can take 1-2 minutes)
Debug panel not appearing (browser cache issue - try hard refresh)

If hotel app fails for an attendee:

Pair them with someone whose app works
Direct them to desktop client alternative
Use Workato’s direct API test as fallback