OllamaSharp support for MCP Resources?

[vrenken]

Hi all,

not an expert here, so whilst skimming through the source code of the OllamaSharp, CSharp ModelContextProtocol and Microsoft.Extensions.AI projects I get somehow confused and think that asking a question here is faster:

Am I correct in the assumption that in the current implementation OllamaSharp does not yet support MCP resources?

My line of reasoning for this:

• I have seen an ASP.NET Core MCP server provide resources, which can be invoked by npx @modelcontextprotocol/inspector,
• I can iterate the resources using the MCP client side API (i.e. using IMcpClient.ListResourcesAsync and IMcpClient.ListResourceTemplatesAsync.
• Where MCP tools need to be assigned to Microsoft.Extensions.AI.ChatOptions.Tools for Ollama (as a Microsoft.Extensions.Ai.IChatClient) to find them, there seems to be no similar construct for resources.

Or is there another concept by which the LLM through OllamaSharp is made aware of the available resources? I would expect these to be somehow assigned to the chat context itself, but am unable to find any code, samples or documentation that describes how this functions.

Thanks in advance & keep up the great work!

Kind regards

Peter Vrenken

[RynFaulkner]

You can use OllamaSharp.ModelContextProtocol, which contains the MCP support functions. From there, if you already have the McpClientTool list from the ModelContextProtocol nuget package which OllamaSharp.ModelContextProtocol is built on, you can use something similar to the code here in Tools.cs, around line 85, to convert them into OllamaSharp's version of the McpClientTool object. From there you can pass the list of tools directly into a chat.SendAsync(string prompt, IEnumberable<object>? tools) call as you would any other prompt.

If you don't already have them, you can use Tools.GetFromMcpServers(McpServerConfiguration[] mcpServers) to get a list you can directly pass into the tools parameter of chat.SendAsync().

For the Microsoft.Extensions.AI, I can't answer that question, as I am only familiar with the ModelContextProtocol nuget support.

[vrenken]

Thanks @RynFaulkner!

We' re looking into it. But as we are not SME' s just a quick validation: Are MCP 'Tools' and 'Resources' really so interoperable?

I.e. in the ModelContextProtocol ecosystem they are described as distinctively different concepts. Surely they are technoligical similar (i.e. based on REST / JSON Schema' s), but does that also mean that they can be mixed and mingled? My rough interpretation was that resources are more 'mandatory' and expand the the context to an LLM - even in a dynamic way, whereas tools are available to the LLM 'when it wishes to use them'. Having 'parameterized' resource URI's with change notification like specified here feels like a very strong concept.

Apologies if this does not make sense. Just trying to find my bearing.

[vrenken]

you can use something similar to the code here in Tools.cs, around line 85, to convert them into OllamaSharp's version of the McpClientTool object.

Also we just checked: The OllamaSharp McpClientTool requires a MCP McpClientTool to be constructed, but the latter sadly only has an internal constructor. So that route would only be accessible with some adventurous reflection - which feels like the wrong aprpoach.

[silentboy3dx]

I follow this discussion with great interest. As you guys are integrating the whole protocol i am learning the concepts as well. I have my tools integrated next up if provide resources. I will wait patiently ..

[awaescher]

Right now, this project has limited support for MCP, which is tools only. I am not planning to extend this at the moment. The reason is that OllamaSharp fully implements IChatClient from Microsoft.Extensions.AI.
By embracing this concept, you'll get the full package of extensions for your AI client, like full MCP support, caching, logging, error handling etc. You can also use Ollama with .NET Aspire and Semantic Kernel with all the goodies you'll get there.
While doing this, OllamaSharp is pulling the strings under the hood and doing the heavy work behind the scenes to communicate with your Ollama instance.

Using OllamaSharp directly is great if you need the proprietary Ollama API functions and for a quick chat implementation with the Chatclass.
For everything else, use OllamaSharp behind the abstraction of IChatClient.

[awaescher]

Resources:

• Usage with Microsoft.Extensions.AI
• Minimalistic demo by eusebio