SDK fails to deserialize agent tool definitions, causing 500 error

[anya2975]

Issue: Azure AI SDK v1.2.0-beta.5 fails to deserialize agent tool definitions when calling GetAIAgentAsync(), causing 500 errors in applications trying to load agent metadata.

Exact Error Details
Stack Trace:
at OpenAI.Responses.InternalUnknownTool.System.ClientModel.Primitives.IPersistableModel<OpenAI.Responses.ResponseTool>.Create(BinaryData data, ModelReaderWriterOptions options)
at System.ClientModel.Primitives.ModelReaderWriter.ReadInternal(BinaryData data, Type returnType, ModelReaderWriterOptions options, ModelReaderWriterContext context)
at System.ClientModel.Primitives.ModelReaderWriter.ReadInternal[T](BinaryData data, ModelReaderWriterOptions options, ModelReaderWriterContext context)
at System.ClientModel.Primitives.ModelReaderWriter.Read[T](BinaryData data, ModelReaderWriterOptions options, ModelReaderWriterContext context)
at Azure.AI.Projects.OpenAI.PromptAgentDefinition.DeserializeToolsValue(JsonProperty property, IList1& tools) at Azure.AI.Projects.OpenAI.PromptAgentDefinition.DeserializePromptAgentDefinition(JsonElement element, ModelReaderWriterOptions options) at Azure.AI.Projects.OpenAI.AgentDefinition.DeserializeAgentDefinition(JsonElement element, ModelReaderWriterOptions options) at Azure.AI.Projects.OpenAI.AgentVersion.DeserializeAgentVersion(JsonElement element, ModelReaderWriterOptions options) at Azure.AI.Projects.OpenAI.AgentObjectVersions.DeserializeAgentObjectVersions(JsonElement element, ModelReaderWriterOptions options) at Azure.AI.Projects.OpenAI.AgentRecord.DeserializeAgentRecord(JsonElement element, ModelReaderWriterOptions options) at Azure.AI.Projects.OpenAI.AgentRecord.PersistableModelCreateCore(BinaryData data, ModelReaderWriterOptions options) at Azure.AI.Projects.OpenAI.AgentRecord.System.ClientModel.Primitives.IPersistableModel<Azure.AI.Projects.OpenAI.AgentRecord>.Create(BinaryData data, ModelReaderWriterOptions options) at System.ClientModel.Primitives.ModelReaderWriter.ReadInternal(BinaryData data, Type returnType, ModelReaderWriterOptions options, ModelReaderWriterContext context) at System.ClientModel.Primitives.ModelReaderWriter.ReadInternal[T](BinaryData data, ModelReaderWriterOptions options, ModelReaderWriterContext context) at System.ClientModel.Primitives.ModelReaderWriter.Read[T](BinaryData data, ModelReaderWriterOptions options, ModelReaderWriterContext context) at Azure.AI.Projects.AzureAIProjectChatClientExtensions.GetAgentRecordByNameAsync(AIProjectClient aiProjectClient, String agentName, CancellationToken cancellationToken) at Azure.AI.Projects.AzureAIProjectChatClientExtensions.GetAIAgentAsync(AIProjectClient aiProjectClient, String name, IList1 tools, Func`2 clientFactory, IServiceProvider services, CancellationToken cancellationToken)

Key Information for Bug Report
SDK Version: Azure.AI.Projects SDK v1.2.0-beta.5
Method: GetAIAgentAsync()
Failing Component: PromptAgentDefinition.DeserializeToolsValue()
Root Cause: SDK cannot deserialize certain tool definitions in agent configurations
Impact: Applications cannot load agent metadata, causing 500 errors
Workaround: Catch deserialization errors and return basic metadata without tool parsing

Environment
Agent: court-listener-helper-2
Framework: ASP.NET Core 9 with Azure AI Foundry integration
The agent exists and works for chat/streaming operations, but the SDK fails when trying to parse its tool configuration for metadata display.

Agent works in playground and the preview, not in this deployed application.

kind: prompt
model: gpt-5-mini
instructions: custom
tools:

• type: mcp
  server_label: court_listener_mcp_v2
  server_url: https://xxxx.northcentralus.azurecontainerapps.io/mcp/
  allowed_tools:
  • smart_opinions_search
  • smart_get_opinion
  • smart_get_cluster
  • smart_get_opinion
  • smart_get_cluster
  • search_opinions
  • search_dockets
  • search_dockets_with_documents
  • search_recap_documents
  • search_audio
  • search_people
  • get_opinion
  • get_docket
  • get_audio
  • get_court
  • get_person
  • get_cluster
  • lookup_citation
  • batch_lookup_citations
  • verify_citation_format
  • parse_citation_with_citeurl
  • extract_citations_from_text
  • enhanced_citation_lookup
  • list_titles
  • list_agencies
  • search_regulations
  • list_all_corrections
  • list_corrections_by_title
  • get_search_suggestions
  • get_search_summary
  • get_title_search_counts
  • get_daily_search_counts
  • get_ancestry
  • get_title_structure
  • get_source_xml
  • get_source_json
  • status
  • get_api_status
  • health_check
    require_approval: never
    project_connection_id: court-listener-mcp-v2

[anya2975]

Also note that the starter code in foundry for the agent is a different approach than what's in this app.

// dotnet add package Azure.AI.Projects --version 1.2.-
using Azure.AI.Projects;
using Azure.AI.Projects.OpenAI;
using Azure.Identity;
using OpenAI;
using OpenAI.Responses;

#pragma warning disable OPENAI001

const string projectEndpoint = "xx";
const string agentName = "xx";

// Connect to your project using the endpoint from your project page
// The AzureCliCredential will use your logged-in Azure CLI identity, make sure to run az login first
AIProjectClient projectClient = new(endpoint: new Uri(projectEndpoint), tokenProvider: new DefaultAzureCredential());

AgentRecord agentRecord = projectClient.Agents.GetAgent(agentName);
Console.WriteLine($"Agent retrieved (name: {agentRecord.Name}, id: {agentRecord.Id})");

OpenAIResponseClient responseClient = projectClient.OpenAI.GetProjectResponsesClientForAgent(agentRecord);
// Use the agent to generate a response
OpenAIResponse response = responseClient.CreateResponse("Hello! Tell me a joke.");

Console.WriteLine(response.GetOutputText());