🇫🇷 Version française

[!IMPORTANT] Every fixture, rulebook, and calculator output in this lab is synthetic and non-binding. Nothing here represents an actual Desjardins product, rate, or policy, and no regulator or insurer has reviewed or endorsed this material.

Overview

Item	Value
Duration	25 minutes
Level	Intermediate
Prerequisites	Lab 00

Learning Objectives

By the end of this lab, you will be able to:

• Explain why mcp/application-server runs independently of apps/workshop and the agent, with no shared runtime
• Confirm the server exposes exactly one tool, get_application(fixture_id)
• Start the server locally and query it for a known and an unknown fixture ID
• Confirm the server never returns write-capable behavior of any kind

Exercises

Exercise 4.1: Read the Server

Open mcp/application-server/main.py and mcp/application-server/application_data_loader.py. Note that get_application looks up a fixture by ID from the data loaded at startup and returns an explicit {"fixtureId": ..., "error": "fixture not found"} object for any ID it does not recognize, rather than raising an exception or returning an empty result.

Exercise 4.2: Run the Existing Tests

python -m pytest mcp/application-server/tests -v

Expected result: every test passes, including cases for a known fixture, an unknown fixture ID, and a check that no other tool is registered on this server.

Exercise 4.3 (Hands-on): Start the Server and Query It

In one terminal, start the server:

python mcp/application-server/main.py

The server listens on http://0.0.0.0:8001 using the streamable-http transport by default. Leave it running, then in a second terminal, call it over MCP the same way the agent does:

python -c "
import asyncio
from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async def main():
    async with streamablehttp_client('http://localhost:8001/mcp') as (read, write, _):
        async with ClientSession(read, write) as session:
            await session.initialize()
            tools = await session.list_tools()
            print('tools:', [t.name for t in tools.tools])
            for fixture_id in ('CASE-SYN-001', 'CASE-SYN-999'):
                result = await session.call_tool('get_application', {'fixture_id': fixture_id})
                print(fixture_id, '->', result.content[0].text)

asyncio.run(main())
"

Expected result in the second terminal: tools: ['get_application'], then the known fixture ID returns the fixture’s input content and the unknown ID returns an explicit error field, never a raised exception or a fabricated fixture.

Switch back to the first terminal. Each call shows up in the server console:

Processing request of type ListToolsRequest
Processing request of type CallToolRequest
get_application('CASE-SYN-001') -> found
Processing request of type CallToolRequest
get_application('CASE-SYN-999') -> fixture not found

Stop the server with Ctrl+C when you are done.

Validation Checklist

• pytest mcp/application-server/tests -v passes
• A known fixtureId returns the exact fixture input
• An unknown fixtureId returns an explicit error result, not an exception
• You confirmed the server exposes no tool other than get_application

Knowledge Check

• Why does this server load its data once at startup rather than reading from disk on every call?
• What would go wrong if get_application raised an exception for an unknown fixture ID instead of returning an error object?

Next Steps

Continue to Lab 05: The Rulebook MCP Server.