Image generated by OpenArt

This is part 5 in our article series about adding AI superpower to your Delphi & C++Builder apps. In the first 4 installments, we covered:

• Introduction to using LLMs via REST API
• Function calling from LLMs
• Multimodal LLM usage
• MCP servers

In this article, we continue our journey along the path of the Model Context Protocol (MCP). This time, our focus is on MCP clients — what they are, how they work, how you can use them and how to create your own. If you haven’t already, we recommend reading our previous article to get a basic understanding of MCP.

What is an MCP Client and Host?

Strictly speaking, an MCP client is a protocol client implementation that maintains a one-to-one connection with an MCP server, regardless of the underlying protocol or connection type used. The sole purpose of the MCP client is to communicate with the server. It provides an interface for sending and receiving data over an active connection.

While that might sound like limited functionality, it’s a critical building block for creating something more powerful and versatile: MCP host applications!

An MCP host is where the real magic happens — it uses MCP clients internally to connect to MCP servers and acts as the glue between the MCP servers and the large language model (LLM). When a prompt is sent to the LLM, the host sends along any available tools for function calling. If the LLM decides it needs to use one of those tools, it includes that in its response. The MCP host then forwards the function parameters to the appropriate MCP server and waits for the result. Once received, the host delivers the response back to the LLM, allowing it to proceed with generating the final answer or to call another tool if needed.

Note: We often refer to MCP host applications as MCP client applications, as this aligns better with Delphi development terminology.

Which MCP Client to Use? How to Set It Up?

There are several ready-to-use MCP client applications available today that can get you up and running quickly. Some popular options include:

• Claude Desktop
• Visual Studio Code
• Cursor.ai

Each of these applications come with a configuration file (usually in JSON format) where you can define connections to one or more MCP servers. Configuration typically involves specifying the server's command, arguments, and possibly authentication tokens or endpoint details.

Here's a general overview of what an MCP host configuration might look like:

{
  "servers": [
    {
      "name": "my-server",
      "command": "path/to/my_server",
      "args": ["--flag1", "value"]
    }
  ]
}

These hosts are excellent for prototyping or integrating LLM functionality into your daily workflows.

But what if you want to embed this power directly into your own Delphi application?

Introducing TTMSMCPClient

We created TTMSMCPClient to make building MCP client applications accessible to Delphi developers.

Looking at the component name, you might wonder: Do I have to handle all communication with the LLM myself? What does TTMSMCPClient do for me?

TTMSMCPClient is not just a barebones MCP client implementation. It allows you to add multiple MCP server connections using the Servers property, each handled by its own internal MCP client. You can configure these servers at design-time or programmatically:

//Add your own custom server
procedure TForm1.FormCreate(Sender: TObject);
var
  server: TTMSMCPClientServerItem;
begin
  server := TMSMCPClient1.Servers.Add;
  server.Command := 'path/to/my_server';
  server.Args.Add('myArg');
  server.Start;
end;

//Add an existing 3rd party server
//https://mcp.so/server/playwright-mcp/microsoft
procedure TForm1.FormCreate(Sender: TObject);
var
  server: TTMSMCPClientServerItem;
begin
  server := TMSMCPClient1.Servers.Add;
  server.Command := 'npx';
  server.Args.Add('@playwright/mcp@latest');
  server.Start;
end;

We went a step further and built in communication with LLMs via our TTMSFNCCloudAI component, automating the full exchange between MCP servers and the LLM. To send a prompt, you only need a single line of code:

procedure TForm1.Button1Click(Sender: TObject);
begin
  TMSMCPClient1.Execute('What is the weather today in Barcelona?');
end;

Capturing the response is just as simple:

procedure TForm1.TMSMCPClient1Executed(Sender: TObject; AResponse: string;
  AHttpStatusCode: Integer; AHttpResult: string);
begin
  if AResponse <> '' then
    ShowMessage(AResponse);
end;

Flexibility and Control

While TTMSMCPClient shares many responsibilities typically handled by an MCP host, it's important to clarify that it is not a host itself. It's a building block that developers can use to create such applications. Therefore we designed it to be as flexible as possible:

• ToolCallMode property lets you enable or disable tool call confirmation (great for automated or background clients)
• Works in both GUI and console applications
• Full control via events — detect what’s happening and act accordingly
• Easily switch to another supported LLM

Use Cases

Using TTMSMCPClient in Delphi applications can open up new possibilities:

• Integrate your own or third-party MCP servers directly into Delphi applications
• Handle LLM responses natively without relying on external tools
• Build full-featured AI-powered applications with customized UI and interaction logic

Conclusion

Applications that use MCP clients can leverage the power of existing or custom MCP servers to unlock the full capabilities of LLMs — now directly within Delphi or C++Builder environments!

In the next article in this series, we’ll dive deeper into using RAG.

If you have an active TMS ALL-ACCESS license, you can now get also access to the first test version of TMS AI Studio that uses the TTMSFNCCloudAI component but also has everything on board to let you build MCP servers and clients. 

Register now to participate in this testing via this landing page.