Version 2025.10.0 of sgcWebSockets expands the AI integration toolkit with a commercial-grade MCP Client implementation that keeps your Delphi applications in lockstep with the Model Context Protocol 2025-06-18 specification. Deliver curated context, orchestrate tool calls and serve resource-aware assistants, all from a single component. 

The Model Context Protocol (MCP) standardizes how assistants negotiate capabilities and exchange structured context. By adding MCP Client support, sgcWebSockets enables:

  • Enterprise-grade interoperability with any MCP-compliant assistant, ensuring consistent initialization and capability discovery workflows.
  • Faster time to market for AI-augmented experiences in existing VCL, FMX or WebBroker projects thanks to a drop-in Delphi component.
  • Operational visibility across every JSON-RPC exchange through built-in HTTP logging and TLS 1.3 transport hardening.

Highlights of TsgcAI_MCP_Client

Turnkey MCP Handshake 

Execute the protocol negotiation from a single Initialize call. The client advertises your product name, title and semantic version while honoring the session identifier returned by the server.

Rich Capability Surface

Issue ready-made method calls including Ping, ToolsList, ToolsCall, PromptsList, PromptsGet, ResourcesList and ResourcesRead. Every response type is strongly typed, making downstream processing effortless.

Event-driven Customization

Hook into granular events—initialization, ping, tool discovery, prompt retrieval and resource streaming—to trace and personalize every exchange before it reaches your business layer.


The component automatically increments JSON-RPC request identifiers, persists MCP session identifiers between calls, and raises typed exceptions when a remote error is returned. HTTP connectivity is encapsulated inside the specialized client that logs traffic and negotiates TLS 1.3 with OpenSSL 3.0 APIs.

Configuration Overview

  • Client Profile: Configure MCPOptions.ClientInfo to present your assistant-integrated solution with name, title and version metadata.
  • Server Endpoint: Set MCPOptions.ServerOptions.URL to the MCP-compatible HTTPS endpoint you want to reach.
  • HTTP Stack: Built on TsgcAI_MCP_HTTP_Client, the transport layer is tuned for JSON content types, automatic session headers and secure TLS defaults.
  • Observability Hooks: Subscribe to event handlers such as OnMCPToolsCall or OnMCPResourcesRead to audit and enrich every protocol transition.

Delphi Integration Example

The snippet below illustrates how to drop the MCP Client onto a data module, initialize it during startup and serve a tool call response as soon as the remote assistant requests it. 

uses
  sgcAI_MCP_Client, sgcJSON;

procedure TdmMCP.StartMCP;
begin
  sgcMCP := TsgcAI_MCP_Client.Create(Self);
  sgcMCP.MCPOptions.ClientInfo.Name := 'sgc-demo-pos';
  sgcMCP.MCPOptions.ClientInfo.Title := 'Smart POS Assistant';
  sgcMCP.MCPOptions.ClientInfo.Version := '2025.10.0';
  sgcMCP.MCPOptions.ServerOptions.URL := 'https://mcp.partnercloud.com';

  sgcMCP.OnMCPInitialize := DoMCPInitialize;
  sgcMCP.OnMCPToolsCall := DoMCPToolsCall;
  sgcMCP.Initialize;
end;

procedure TdmMCP.DoMCPInitialize(Sender: TObject;
  const Request: TsgcAI_MCP_Request_Initialize;
  const Response: TsgcAI_MCP_Response_Initialize;
  var Accept: Boolean);
begin
  Accept := Response.ServerInfo.SupportsTools('inventory.lookup');
end;

procedure TdmMCP.DoMCPToolsCall(Sender: TObject;
  const Request: TsgcAI_MCP_Request_ToolsCall;
  const Response: TsgcAI_MCP_Response_ToolsCall);
var
  ResultPayload: IsgcJSON;
begin
  if Request.Params.Name = 'inventory.lookup' then
  begin
    ResultPayload := TsgcJSON.CreateObject;
    ResultPayload['sku'] := Request.Params.Arguments['sku'];
    ResultPayload['availability'] := 'in-stock';
    Response.Result := ResultPayload;
  end;
end;

Delphi Sample

sgcMCPClient
4.9 mb
Download Files