[QUESTION] ChatTools and ResponseTools helper classes missing?

[ecarter-gny]

I might be going crazy here, but I cannot find these helper classes anywhere in the 2.2.0 release even though they are indicated in the release notes.

Re:
Add ChatTools and ResponseTools helper classes by @christothes in #422

[jsquire]

Hi @ecarter-gny. The MCP tools helper classes from that pull request were exploratory and have not been released. That pull request spawned additional discussion about what form they should take, which package they should live in, and alignment with service features which have still not concluded. At present, we do not have plans to move forward with them in the form they appeared.

There shouldn't be any mention of them in release notes - if there are, then it was a mistake. I scanned the official change log and I'm not seeing a reference to them. Can you help me understand where you saw them appear?

[ecarter-gny]

Referring to the release notes here: https://github.com/openai/openai-dotnet/releases/tag/OpenAI_2.2.0

The first line item.

I was hoping to use the helper utilities to add mcp tool functions for use in the responses API.

[jsquire]

Thanks, @ecarter-gny, that was helpful. It looks like we missed a step in our release workflow that would point these at the changelog and GitHub was generating default content. Because features currently flow from an upstream repository into this one, the default content is misleading. The initial exploratory MCP PR took place here, but the revert happened upstream and flowed down in one of the large "release prep" pull requests from last week, making it difficult to identify.

We apologize for the confusion. I've opened #526 to address this so that we have one consistent set of documentation for changes in a release.

[ecarter-gny]

Great, thank you!

Do you have any information to provide regarding adding remote MCP servers as tools to calls to the Responses API? It doesn't seem to be supported at the moment, and I haven't really been able to find any help online with accomplishing this.

This thread indicates this functionality as (Implemented and merged): #406

#406 (comment)

[jsquire]

@ecarter-gny: It is not supported for the reasons that I mentioned above. In short, conversations about how to do this in a way that best aligns with the OpenAI service plans and the overall ecosystem are ongoing. We apologize that the discussion thread in that issue is misleading. I'll leave a comment there to clarify.

In the meantime, the suggestion from Stephen in the thread is the best path forward at the moment. For convenience, reflecting that here:

Note you can already do this using the OpenAI library in conjunction with the official C# MCP SDK and Microsoft.Extensions.AI, e.g.

using Microsoft.Extensions.AI;
using ModelContextProtocol.Client;
using ModelContextProtocol.Protocol.Transport;
using OpenAI;

var chatClient = new OpenAIClient(Environment.GetEnvironmentVariable("AI:OpenAI:ApiKey"))
    .GetOpenAIResponseClient("o3-mini")
    .AsIChatClient()
    .AsBuilder()
    .UseFunctionInvocation()
    .Build();

await using var mcpClient = await McpClientFactory.CreateAsync(new StdioClientTransport(new()
{
    Command = "npx",
    Arguments = ["-y", "@modelcontextprotocol/server-everything"],
}));

var tools = await mcpClient.ListToolsAsync();
foreach (var tool in tools)
{
    Console.WriteLine($"Tool: {tool.Name}, Description: {tool.Description}");
}

Console.WriteLine(await chatClient.GetResponseAsync("Use the echo tool to echo 'Hello world'", new() { Tools = [.. tools] }));

This prints:

Tool: echo, Description: Echoes back the input
Tool: add, Description: Adds two numbers
Tool: printEnv, Description: Prints all environment variables, helpful for debugging MCP server configuration
Tool: longRunningOperation, Description: Demonstrates a long running operation with progress updates
Tool: sampleLLM, Description: Samples from an LLM using MCP's sampling feature
Tool: getTinyImage, Description: Returns the MCP_TINY_IMAGE
Tool: annotatedMessage, Description: Demonstrates how annotations can be used to provide metadata about content
Tool: getResourceReference, Description: Returns a resource reference that can be used by MCP clients
Echo: Hello world

[KrzysztofCwalina]

More complete sample showing how to call MCP with OpenAI CHatClient can be found at https://github.com/openai/openai-dotnet/blob/main/docs/guides/mcp/chat_with_mcp.cs

[ecarter-gny]

Thanks all, I was able to make some progress on this. It's not the most intuitive approach, hopefully we can get something within the openai-dotnet library to make it simpler to support it through the Responses API. I had to essentially refactor my code to switch from using the OpenAI libraries to using the Microsoft.Extensions.AI ChatMessage/ChatTool/ChatOptions library which then made it harder to support filesearch with MCP.

This could've been just a skill issue on my part though.

The sample @KrzysztofCwalina just linked was very useful though, thank you.