Howie/samples for tests

[howieleung]

This is a POC to be extended for all tests for tools.

1. some sample such as this AI Search sample accept user input. Now they can be skip if you specify input by env var.
2. use try catch finally with finally for tear down
3. assert output at the end.

[Copilot]

Pull request overview

This PR introduces improvements to make samples more testable by adding environment variable support for user input, implementing proper try-catch-finally cleanup patterns, and adding assertions to validate outputs. The PR also reorganizes MCP sample files by moving sample_mcp_tool_async.py to a new mcp_client directory with supporting assets.

Key changes:

• Added optional environment variable AI_SEARCH_USER_INPUT to skip interactive prompts in the AI Search sample
• Implemented try-finally pattern for reliable agent cleanup in sample_agent_ai_search.py
• Added output assertion to validate sample execution success
• Relocated MCP tool sample to dedicated mcp_client directory with improved documentation

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
sdk/ai/azure-ai-projects/samples/mcp_client/sample_mcp_tool_async.py	New location for MCP sample with enhanced documentation and workflow description
sdk/ai/azure-ai-projects/samples/mcp_client/assets/product_info.md	Asset file for file search demonstration in MCP sample
sdk/ai/azure-ai-projects/samples/agents/tools/sample_agent_ai_search.py	Refactored with main() function, environment variable support, try-finally cleanup, and output assertions
sdk/ai/azure-ai-projects/samples/agents/sample_mcp_tool_async_.py	Removed old sample location (appears to be deleted/moved)
sdk/ai/azure-ai-projects/CHANGELOG.md	Updated to document the improved MCP client sample

[Copilot]

The file handle opened on line 123 is never closed. While the async with context manager will clean up the session, the file opened with open() should be properly closed to prevent resource leaks.

Consider using a context manager:

with open(
    os.path.abspath(os.path.join(os.path.dirname(__file__), "./assets/product_info.md")),
    "rb",
) as f:
    vector_store_file = await openai_client.vector_stores.files.upload_and_poll(
        vector_store_id=vector_store.id,
        file=f,
    )

Suggested change

	vector_store_file = await openai_client.vector_stores.files.upload_and_poll(
	vector_store_id=vector_store.id,
	file=open(
	os.path.abspath(os.path.join(os.path.dirname(__file__), "./assets/product_info.md")),
	"rb",
	),
	)
	with open(
	os.path.abspath(os.path.join(os.path.dirname(__file__), "./assets/product_info.md")),
	"rb",
	) as f:
	vector_store_file = await openai_client.vector_stores.files.upload_and_poll(
	vector_store_id=vector_store.id,
	file=f,
	)

[Copilot]

The sample creates a vector store but never cleans it up. This can lead to resource accumulation over time. Consider adding cleanup in a try-finally block to ensure the vector store is deleted even if an error occurs, similar to how other samples handle agent cleanup.

Example pattern:

vector_store = None
try:
    vector_store = await openai_client.vector_stores.create(...)
    # ... use vector store ...
finally:
    if vector_store:
        await openai_client.vector_stores.delete(vector_store_id=vector_store.id)

Suggested change

	# Create a vector store
	vector_store = await openai_client.vector_stores.create(
	name="sample_vector_store",
	)
	vector_store_file = await openai_client.vector_stores.files.upload_and_poll(
	vector_store_id=vector_store.id,
	file=open(
	os.path.abspath(os.path.join(os.path.dirname(__file__), "./assets/product_info.md")),
	"rb",
	),
	)
	print(f"\n\nUploaded file, file ID: {vector_store_file.id} to vector store ID: {vector_store.id}")
	# Call the file_search tool
	file_search_result = await session.call_tool(
	name="file_search",
	arguments={"queries": ["What feature does Smart Eyewear offer?"]},
	meta={"vector_store_ids": [vector_store.id]},
	)
	print(f"\n\nFile Search Output: {file_search_result.content}")
	# Create a vector store and ensure cleanup
	vector_store = None
	try:
	vector_store = await openai_client.vector_stores.create(
	name="sample_vector_store",
	)
	vector_store_file = await openai_client.vector_stores.files.upload_and_poll(
	vector_store_id=vector_store.id,
	file=open(
	os.path.abspath(os.path.join(os.path.dirname(__file__), "./assets/product_info.md")),
	"rb",
	),
	)
	print(f"\n\nUploaded file, file ID: {vector_store_file.id} to vector store ID: {vector_store.id}")
	# Call the file_search tool
	file_search_result = await session.call_tool(
	name="file_search",
	arguments={"queries": ["What feature does Smart Eyewear offer?"]},
	meta={"vector_store_ids": [vector_store.id]},
	)
	print(f"\n\nFile Search Output: {file_search_result.content}")
	finally:
	if vector_store:
	await openai_client.vector_stores.delete(vector_store_id=vector_store.id)

[Copilot]

[nitpick] The sample creates a file (puppy.png) but doesn't clean it up afterwards. For a sample that may be run multiple times, consider cleaning up generated artifacts.

You could add cleanup after the sample completes:

try:
    # ... existing code ...
finally:
    # Clean up generated files
    if os.path.exists("puppy.png"):
        os.remove("puppy.png")

Suggested change

	async with (
	DefaultAzureCredential() as credential,
	AIProjectClient(endpoint=endpoint, credential=credential) as project_client,
	project_client.get_openai_client() as openai_client,
	streamablehttp_client(
	url=f"{endpoint}/mcp_tools?api-version=2025-05-15-preview",
	headers={"Authorization": f"Bearer {(await credential.get_token('https://ai.azure.com')).token}"},
	) as (read_stream, write_stream, _),
	ClientSession(read_stream, write_stream) as session,
	):
	# Initialize the connection
	await session.initialize()
	# List available tools
	tools = await session.list_tools()
	print(f"Available tools: {[tool.name for tool in tools.tools]}")
	# For each tool, print its details
	for tool in tools.tools:
	print(f"\n\nTool Name: {tool.name}, Input Schema: {tool.inputSchema}")
	# Run the code interpreter tool
	code_interpreter_result = await session.call_tool(
	name="code_interpreter",
	arguments={"code": "print('Hello from Microsoft Foundry MCP Code Interpreter tool!')"},
	)
	print(f"\n\nCode Interpreter Output: {code_interpreter_result.content}")
	# Run the image_generation tool
	image_generation_result = await session.call_tool(
	name="image_generation",
	arguments={"prompt": "Draw a cute puppy riding a skateboard"},
	meta={"imagegen_model_deployment_name": os.getenv("IMAGE_GEN_DEPLOYMENT_NAME", "")},
	)
	# Save the image generation output to a file
	if image_generation_result.content and isinstance(image_generation_result.content[0], ImageContent):
	filename = "puppy.png"
	file_path = os.path.abspath(filename)
	print(f"\nImage saved to: {file_path}")
	with open(file_path, "wb") as f:
	f.write(base64.b64decode(image_generation_result.content[0].data))
	# Create a vector store
	vector_store = await openai_client.vector_stores.create(
	name="sample_vector_store",
	)
	vector_store_file = await openai_client.vector_stores.files.upload_and_poll(
	vector_store_id=vector_store.id,
	file=open(
	os.path.abspath(os.path.join(os.path.dirname(__file__), "./assets/product_info.md")),
	"rb",
	),
	)
	print(f"\n\nUploaded file, file ID: {vector_store_file.id} to vector store ID: {vector_store.id}")
	# Call the file_search tool
	file_search_result = await session.call_tool(
	name="file_search",
	arguments={"queries": ["What feature does Smart Eyewear offer?"]},
	meta={"vector_store_ids": [vector_store.id]},
	)
	print(f"\n\nFile Search Output: {file_search_result.content}")
	try:
	async with (
	DefaultAzureCredential() as credential,
	AIProjectClient(endpoint=endpoint, credential=credential) as project_client,
	project_client.get_openai_client() as openai_client,
	streamablehttp_client(
	url=f"{endpoint}/mcp_tools?api-version=2025-05-15-preview",
	headers={"Authorization": f"Bearer {(await credential.get_token('https://ai.azure.com')).token}"},
	) as (read_stream, write_stream, _),
	ClientSession(read_stream, write_stream) as session,
	):
	# Initialize the connection
	await session.initialize()
	# List available tools
	tools = await session.list_tools()
	print(f"Available tools: {[tool.name for tool in tools.tools]}")
	# For each tool, print its details
	for tool in tools.tools:
	print(f"\n\nTool Name: {tool.name}, Input Schema: {tool.inputSchema}")
	# Run the code interpreter tool
	code_interpreter_result = await session.call_tool(
	name="code_interpreter",
	arguments={"code": "print('Hello from Microsoft Foundry MCP Code Interpreter tool!')"},
	)
	print(f"\n\nCode Interpreter Output: {code_interpreter_result.content}")
	# Run the image_generation tool
	image_generation_result = await session.call_tool(
	name="image_generation",
	arguments={"prompt": "Draw a cute puppy riding a skateboard"},
	meta={"imagegen_model_deployment_name": os.getenv("IMAGE_GEN_DEPLOYMENT_NAME", "")},
	)
	# Save the image generation output to a file
	if image_generation_result.content and isinstance(image_generation_result.content[0], ImageContent):
	filename = "puppy.png"
	file_path = os.path.abspath(filename)
	print(f"\nImage saved to: {file_path}")
	with open(file_path, "wb") as f:
	f.write(base64.b64decode(image_generation_result.content[0].data))
	# Create a vector store
	vector_store = await openai_client.vector_stores.create(
	name="sample_vector_store",
	)
	vector_store_file = await openai_client.vector_stores.files.upload_and_poll(
	vector_store_id=vector_store.id,
	file=open(
	os.path.abspath(os.path.join(os.path.dirname(__file__), "./assets/product_info.md")),
	"rb",
	),
	)
	print(f"\n\nUploaded file, file ID: {vector_store_file.id} to vector store ID: {vector_store.id}")
	# Call the file_search tool
	file_search_result = await session.call_tool(
	name="file_search",
	arguments={"queries": ["What feature does Smart Eyewear offer?"]},
	meta={"vector_store_ids": [vector_store.id]},
	)
	print(f"\n\nFile Search Output: {file_search_result.content}")
	finally:
	# Clean up generated files
	puppy_file = "puppy.png"
	if os.path.exists(puppy_file):
	os.remove(puppy_file)

[Copilot]

The output variable is assigned to None on line 102 but then conditionally assigned within the event loop. If the loop completes without encountering a response.completed event, output will remain None, causing the assertion on line 138 to fail.

This initialization should be removed since output is already initialized as None at the module level (line 45), and the code should ensure the assertion message is more descriptive about what went wrong.

[Copilot]

The exception is caught and printed but then re-raised, making the error message on line 127 redundant since the exception will propagate anyway with its original message. Consider either:

1. Removing the print statement and just re-raising the exception
2. Logging the error properly instead of printing
3. Not re-raising if you want to handle it gracefully

Example:

except Exception as e:
    # Let the exception propagate with its original traceback
    raise

Suggested change

	print(f"Error occurred: {e}")
	raise e
	raise

[dargilco]

azure-identity non longer needed in this line after my PR goes in.

[dargilco]

Do you need to refresh code snippets in package README.md?