MCPServer Component

Properties   Methods   Events   Config Settings   Errors  

Implements an easy way to expose custom tools, prompts, and resources to Model Context Protocol (MCP) clients.

Syntax

TsdmMCPServer

Remarks

The MCPServer component provides a simple, event-driven API for interacting with clients that use the Model Context Protocol (MCP).

Supported Transport Mechanisms

The MCPServer component supports multiple transport mechanisms that are used to facilitate communication, which can be specified by setting the Transport property.

For local communication scenarios, such as when the server is launched as a subprocess of a client application, the component is capable of communicating via standard input and output streams. This setup is simple and enables direct data exchange between the client and server within the same process environment.

In remote deployment scenarios, where the server must be accessible to multiple concurrent clients across a network, the component also supports HTTP as the transport layer, using standard HTTP requests and responses for communication. To accommodate different hosting needs, the component includes a built-in embedded HTTP server, but it can also be integrated into an existing web infrastructure such as ASP.NET Core, Apache Tomcat, or other frameworks.

While the developer experience is largely the same between these, there are some differences in authentication and request processing depending on how the server is hosted. Comprehensive information on these hosting options and configuration can be found in the Hosting Options page.

Key Features

The MCPServer component supports all major MCP functionality, including tools, prompts, resources, and sampling.

Tools

Tools provide an opportunity for LLM clients to execute custom code, interact with external systems, and access dynamic or external data. Anything for which a function can be written can also be exposed as a tool. Common examples include database queries, file operations, or computations.

Registering Tools

Tools must be registered with the server via the RegisterTool method to be exposed to the client. Each registration requires a unique Name and a natural language Description. The name identifies the tool, while the description helps LLM clients determine when to use it. Since tool selection is client-driven, writing clear and specific descriptions is key to improving tool usage accuracy.

A tool may have zero or more associated parameters. Each parameter must be registered before the tool it will be associated with. Each parameter requires a Name and a Required boolean indicating if that parameter is required.

server.RegisterToolParam("limit", false); server.RegisterTool("random-num", "Generates a random number");

Handling Tool Invocations

Once a tool is registered, the client may invoke it at any time. When this occurs, the ToolRequest event is fired. This event includes the Name of the requested tool, which the server can use to determine how to respond. If the tool requires input, the server can retrieve parameters from the client using the GetToolParamValue method.

If the tool completes successfully and needs to return output, the server may respond using the AddToolMessage method. If an error occurs, such as an unrecognized tool is requested or a failure occurs during execution, the server should set the IsError parameter to True to inform the client that the request failed.

server.OnToolRequest += (s, e) => { if (e.Name == "random-num") { server.AddToolMessage((int)TToolMessageTypes.mtText, randomNum); } else { e.IsError = true; } };

Prompts

Prompts are predefined conversation starters or instructions that can be quickly inserted into the LLM's context to guide specific interactions or workflows. Prompts serve as standardized templates for common conversational patterns and allow for consistent responses for tasks like reviewing code, analyzing data, or answering specific types of questions. Prompts may also include arguments that allow runtime customization.

Unlike tools, prompts are user-driven. The client does not select prompts on its own; instead, the user chooses when and which prompt to insert, which allows for consistent and reusable guidance tailored to specific workflows.

Registering Prompts

Prompts must be registered with the server via the RegisterPrompt method to be exposed to the client. Each registration requires a unique Name and a natural language Description. The name identifies the prompt, while the description helps the user understand its purpose. Once registered, the prompt becomes available for the end user to insert into the client context as needed.

A prompt may have zero or more associated arguments. Each argument must be registered before the prompt it will be associated with. Each argument requires a Name and a Required boolean indicating if that argument is required.

// Define argument schema. server.RegisterPromptArg("code", true); server.RegisterPromptArg("language", false); // Register the prompt. string name = "explain-code"; string description = "Explain how code works"; server.RegisterPrompt(name, description);

Handling Prompt Requests

When a user selects a prompt, the client notifies the server, causing the PromptRequest event to be fired. This event includes the Name of the requested prompt, allowing the server to determine the correct response to be sent using the AddPromptMessage method. Any related prompt arguments may also be retrieved by their associated name using the GetPromptParamValue method.

server.OnPromptRequest += (s, e) => { // Retrieve runtime values sent by client. string code = server.GetPromptParamValue("code"); string language = server.GetPromptParamValue("language") ?? "Unknown"; // Build response messages. // Assuming the client sent 'a = 1 + 2;' for the 'code' argument, and 'python' for the 'language' argument. server.AddPromptMessage((int)TRoles.rtAssistant, "Don't add comments."); server.AddPromptMessage((int)TRoles.rtUser, $"Explain how this {language} code works:\n\n{code}"); };

Resources

Resources provide persistent, read-only content that can be requested once by the user and reused throughout the session. These are typically static files such as documentation, source code, or other reference materials that help provide context for LLM interactions.

Registering Resources

Resources must be registered with the server via the RegisterResource method to be exposed to the client. Each registration requires a unique Uri, a short display Name, and a natural language Description.

server.RegisterResource("file://server.cs", "Server Demo", "The source code for a simple MCP SDK Server Demo");

Handling Resource Requests

When a user requests a resource, the ResourceRequest event is fired. This event includes the Uri of the requested resource, allowing the server to locate and load the correct resource using the AddResourceContent method. The client will manage the resource afterward, making it available for reuse as needed.

server.OnResourceRequest += (s, e) => { string fileName = e.Uri.Substring("file://".Length); if (File.Exists(fileName)) { string content = File.ReadAllText(fileName); server.AddResourceContent(e.Uri, content, "text/plain"); } };

Property List

---

The following is the full list of the properties of the component with short descriptions. Click on the links for further details.

Method List

---

The following is the full list of the methods of the component with short descriptions. Click on the links for further details.

Event List

---

The following is the full list of the events fired by the component with short descriptions. Click on the links for further details.

Config Settings

---

The following is a list of config settings for the component with short descriptions. Click on the links for further details.

Listening Property (MCPServer Component)

Whether the component is listening for incoming connections.

Syntax

property Listening: Boolean read get_Listening;

Default Value

false

Remarks

When the Transport property is set to ttStdio, or it is set to ttHTTP and the ProcessingMode property is set to modeEmbeddedServer, this property indicates whether the component is listening for connections on the port specified by the LocalPort property.

The StartListening and StopListening methods can be used to control whether the component is listening.

This property is read-only.

ProcessingMode Property (MCPServer Component)

The mode in which the component will be hosted.

Syntax

property ProcessingMode: TsdmTMCPProcessingModes read get_ProcessingMode write set_ProcessingMode;

TsdmTMCPProcessingModes = (
  modeEmbeddedServer,
  modeExternalServer,
  modeOffline
);

Default Value

modeEmbeddedServer

Remarks

When the Transport property is set to ttHTTP, this property governs whether the component operates via the embedded HTTP server, an external server, or in a fully-offline mode. Possible values are as follows:

• modeEmbeddedServer (0, default): Uses the embedded server.
• modeExternalServer (1): Uses an external server framework.
• modeOffline (2): Uses offline processing.

Please refer to the Hosting Options pages to learn more about the different processing modes and how they relate to configuring a host server.

Prompts Property (MCPServer Component)

A collection of registered prompts.

Syntax

property Prompts: TsdmPromptList read get_Prompts;

Remarks

This collection holds a list of Prompt items.

Calling the RegisterPrompt and UnregisterPrompt methods will add and remove items to and from this collection, respectively.

This property is read-only and not available at design time.

Please refer to the Prompt type for a complete list of fields.

RegisteredPromptArgs Property (MCPServer Component)

A collection of prompt arguments to be registered.

Syntax

property RegisteredPromptArgs: TsdmPromptArgList read get_RegisteredPromptArgs write set_RegisteredPromptArgs;

Remarks

This collection holds a list of PromptArg items to be registered.

Calling the RegisterPromptArg method will add items to this collection, and calling the RegisterPrompt method will clear this collection.

Example: // Register two prompt argument schemas. server.RegisterPromptArg("code", "Code to explain", true); server.RegisterPromptArg("language", "Programming language", false); // Register the prompt with the added arguments. server.RegisterPrompt("explain-code", "Explain how code works");

This property is not available at design time.

Please refer to the PromptArg type for a complete list of fields.

RegisteredToolParams Property (MCPServer Component)

A collection of tool parameters to be registered.

Syntax

property RegisteredToolParams: TsdmToolParamList read get_RegisteredToolParams write set_RegisteredToolParams;

Remarks

This collection holds a list of ToolParam items to be registered.

Calling the RegisterToolParam method will add items to this collection, and calling the RegisterTool method will clear this collection.

This property is not available at design time.

Please refer to the ToolParam type for a complete list of fields.

Request Property (MCPServer Component)

The body of the HTTP request to be processed by the component.

Syntax

property Request: String read get_Request write set_Request;
property RequestB: TBytes read get_RequestB write set_RequestB;

Default Value

''

Remarks

When the Transport property is set to ttHTTP and the ProcessingMode property is set to modeOffline, the component processes inbound HTTP requests via this property and the RequestHeaders property.

After setting this property to the full body content of the HTTP request, as well as setting any headers in the RequestHeaders property, the component will begin processing the request when the ProcessRequest method is called.

RequestHeaders Property (MCPServer Component)

The full set of headers of the HTTP request to be processed by the component.

Syntax

property RequestHeaders: String read get_RequestHeaders write set_RequestHeaders;

Default Value

''

Remarks

When the Transport property is set to ttHTTP and the ProcessingMode property is set to modeOffline, the component processes inbound HTTP requests via this property and the Request property.

After setting this property to the header content of the HTTP request, as well as setting the body content in the Request property, the component will begin processing the request when the ProcessRequest method is called.

Resources Property (MCPServer Component)

A collection of registered resources.

Syntax

property Resources: TsdmResourceList read get_Resources;

Remarks

This collection holds a list of Resource items.

Calling the RegisterResource and UnregisterResource methods will add and remove items to and from this collection, respectively.

This property is read-only and not available at design time.

Please refer to the Resource type for a complete list of fields.

Response Property (MCPServer Component)

The body of the HTTP response generated after processing a request with the component.

Syntax

property Response: String read get_Response write set_Response;
property ResponseB: TBytes read get_ResponseB write set_ResponseB;

Default Value

''

Remarks

When the Transport property is set to ttHTTP and the ProcessingMode property is set to modeOffline, this property holds the body of the HTTP response that was generated as a result of calling the ProcessRequest method.

Note that this response is not sent to a client, and instead the component simply populates this property along with the ResponseHeaders property.

ResponseHeaders Property (MCPServer Component)

The full set of headers of the HTTP response generated after processing a request with the component.

Syntax

property ResponseHeaders: String read get_ResponseHeaders write set_ResponseHeaders;

Default Value

''

Remarks

When the Transport property is set to ttHTTP and the ProcessingMode property is set to modeOffline, this property holds the headers of the HTTP response that was generated as a result of calling the ProcessRequest method.

Note that this response is not sent to a client, and instead the component simply populates this property along with the Response property.

SamplingMessages Property (MCPServer Component)

A collection of messages to send when requesting language model generation from the client.

Syntax

property SamplingMessages: TsdmSamplingMessageList read get_SamplingMessages write set_SamplingMessages;

Remarks

This collection holds a list of SamplingMessage items.

When the Transport property is set to ttStdio, calling the SendSamplingRequest method sends the messages currently in this collection to the client, where they will be passed into the client's language model as input for text generation.

This property is not available at design time.

Please refer to the SamplingMessage type for a complete list of fields.

ServerCert Property (MCPServer Component)

The certificate used during SSL negotiation.

Syntax

property ServerCert: TsdmCertificate read get_ServerCert write set_ServerCert;

Remarks

When the Transport property is set to ttHTTP and the ProcessingMode property is set to modeEmbeddedServer, this property holds the digital certificate that the component will use during SSL negotiation.

This is the server's certificate, and it must be set before the StartListening method is called.

Please refer to the Certificate type for a complete list of fields.

ServerSettings Property (MCPServer Component)

The embedded HTTP server settings used by the component.

Syntax

property ServerSettings: TsdmServerSettings read get_ServerSettings write set_ServerSettings;

Remarks

When the Transport property is set to ttHTTP and the ProcessingMode property is set to modeEmbeddedServer, this property is used to define the necessary properties to configure the embedded HTTP server.

Please refer to the ServerSettings type for a complete list of fields.

SystemPrompt Property (MCPServer Component)

An instruction sent to the client's language model to influence its behavior during a sampling request.

Syntax

property SystemPrompt: String read get_SystemPrompt write set_SystemPrompt;

Default Value

''

Remarks

When the Transport property is set to ttStdio, this property specifies an instruction sent to the client's language model to influence its behavior when the SendSamplingRequest method is called. It acts as high-level guidance or context for how the language model should interpret and respond to the messages in the SamplingMessages properties.

Example: server.SystemPrompt = "You are an assistant meant to summarize text only using a formal tone."; SamplingMessage message = new SamplingMessage(); message.Role = TRoles.rtUser; message.Text = "Summarize the following text: " + text; server.SamplingMessages.Add(message); string result = server.SendSamplingRequest();

Tools Property (MCPServer Component)

A collection of registered tools.

Syntax

property Tools: TsdmToolList read get_Tools;

Remarks

This collection holds a list of Tool items.

Calling the RegisterTool and UnregisterTool methods will add and remove items to and from this collection, respectively.

This property is read-only and not available at design time.

Please refer to the Tool type for a complete list of fields.

Transport Property (MCPServer Component)

The transport mechanism used for communication.

Syntax

property Transport: TsdmTTransports read get_Transport write set_Transport;

TsdmTTransports = (
  ttStdio,
  ttHTTP
);

Default Value

ttStdio

Remarks

This property indicates whether the component operates via the standard input/output or HTTP transport mechanism. Possible values are as follows:

• ttStdio (1, default): Enables communication through standard input and output streams, but the component is limited to serving a single client at once due to the server acting as a subprocess of a client application.
• ttHTTP (2): Enables communication via HTTP requests and responses, and allows the component to serve multiple clients at once.

This property is not available at design time.

AddPromptMessage Method (MCPServer Component)

Adds a prompt message to be returned during a prompt request.

Syntax

procedure AddPromptMessage(role: Integer; text: String);

Remarks

This method is used to add a new prompt message to be passed back to the client during an ongoing prompt request. It may only be called from within the PromptRequest event.

Role determines the message's origin and context within the conversation flow. Valid roles include:

Text contains natural language describing the prompt that will be passed into the client's language model (e.g., Review python code, Summarize an article, or Provide an example of using the /n software MCP SDK).

Example: server.OnPromptRequest += (s, e) => { if (e.Name.Equals("Greeting")) { server.AddPromptMessage((int)TRoles.rtUser, "Hello!"); } };

Advanced Example: server.OnPromptRequest += (s, e) => { if (e.Name.Equals("Code Review")) { string code = server.GetPromptParamValue("code"); string language = server.GetPromptParamValue("language"); string reviewType = server.GetPromptParamValue("reviewType"); string instructions = "Please review the following " + language + " code."; if (!string.IsNullOrEmpty(reviewType)) { instructions += " focusing on " + reviewType; } instructions += ":"; server.AddPromptMessage((int)TRoles.rtUser, instructions); server.AddPromptMessage((int)TRoles.rtUser, code); } };

AddResourceContent Method (MCPServer Component)

Adds data to be returned during a resource request.

Syntax

procedure AddResourceContent(uri: String; text: String; mimeType: String);

Remarks

This method is used to add resource content to be passed back to the client during an ongoing resource request, providing the actual data that clients are requesting when they access a specific resource. It may only be called from within the ResourceRequest event and may be called multiple times during the same resource request to provide multiple related resources or different versions of the same content.

Uri specifies the unique resource identifier that corresponds to the content being provided.

Text contains text data that will be included in the response for the ongoing resource request.

MimeType (e.g., text/plain or application/json) informs the client of how the data passed into Text should be interpreted and processed.

Example: server.OnResourceRequest += (s, e) => { string uri = e.Uri.ToString(); if (uri.StartsWith("file:///")) { string fileName = uri.Substring("file:///".Length).Replace('/', Path.DirectorySeparatorChar); string fullPath = Path.Combine(baseDir, fileName); if (File.Exists(fullPath)) { string data = File.ReadAllText(fullPath); server.AddResourceContent(uri, data, "text/plain"); } } };

Advanced Example server.OnResourceRequest += (s, e) => { string uri = e.Uri.ToString(); if (uri.StartsWith("file:///")) { string fileName = uri.Substring("file:///".Length).Replace('/', Path.DirectorySeparatorChar); string fullPath = Path.Combine(baseDir, fileName); string fileExtension = Path.GetExtension(fullPath).ToLower(); string mimeType = GetMimeType(fileExtension); if (IsImageFile(fileExtension)) { // Handle image files. byte[] imageBytes = File.ReadAllBytes(fullPath); string base64Image = Convert.ToBase64String(imageBytes); server.AddResourceContent(uri, base64Image, mimeType); } else { // Handle text files. string data = File.ReadAllText(fullPath); server.AddResourceContent(uri, data, mimeType); } } };

AddToolMessage Method (MCPServer Component)

Adds a tool message to be returned during a tool request.

Syntax

procedure AddToolMessage(messageType: Integer; value: String);

Remarks

This method is used to add a new tool message to be passed back to the client during an ongoing tool request. It may only be called from within the ToolRequest event, and should only be called once per each type of data to be returned.

MessageType identifies the data type of the message being added. Valid data types include:

Value contains text that will be included in the response for the ongoing tool request.

Config Method (MCPServer Component)

Sets or retrieves a configuration setting.

Syntax

function Config(ConfigurationString: String): String;

Remarks

Config is a generic method available in every component. It is used to set and retrieve configuration settings for the component.

These settings are similar in functionality to properties, but they are rarely used. In order to avoid "polluting" the property namespace of the component, access to these internal properties is provided through the Config method.

To set a configuration setting named PROPERTY, you must call Config("PROPERTY=VALUE"), where VALUE is the value of the setting expressed as a string. For boolean values, use the strings "True", "False", "0", "1", "Yes", or "No" (case does not matter).

To read (query) the value of a configuration setting, you must call Config("PROPERTY"). The value will be returned as a string.

DoEvents Method (MCPServer Component)

Processes events from the internal queue.

Syntax

procedure DoEvents();

Remarks

This method causes the component to process any available events. If no events are available, it waits for a preset period of time and then returns.

GetPromptParamValue Method (MCPServer Component)

Retrieves the value of an existing prompt parameter.

Syntax

function GetPromptParamValue(Name: String): String;

Remarks

This method is used to retrieve the value of an existing prompt parameter. It may only be called from within the PromptRequest event.

Name contains a unique identifier for the prompt parameter that is being retrieved. This must match the value passed into the RegisterPromptArg method when the parameter was registered.

The value associated with the prompt parameter is returned when calling this method. If the specified parameter does not exist or was not provided by the client, then an empty string is returned.

Example: // Define argument schema. server.RegisterPromptArg("code", "Code to explain", true); server.RegisterPromptArg("language", "Programming language", false); // Register the prompt. string name = "explain-code"; string description = "Explain how code works"; server.RegisterPrompt(name, description); // Handling the prompt request. server.OnPromptRequest += (s, e) => { // Retrieve runtime values sent by client. string code = server.GetPromptParamValue("code"); string language = server.GetPromptParamValue("language") ?? "Unknown"; // Build response messages. // Assuming the client sent 'a = 1 + 2;' for the 'code' argument, and 'python' for the 'language' argument. // The following response should contain a user message with the text 'Explain how this python code works: \n\na = 1 + 2;'. server.AddPromptMessage("user", $"Explain how this {language} code works:\n\n{code}"); };

GetToolParamValue Method (MCPServer Component)

Retrieves the value of an existing tool parameter.

Syntax

function GetToolParamValue(Name: String): String;

Remarks

This method is used to retrieve the value of an existing tool parameter. It may only be called from within the ToolRequest event.

Name contains a unique identifier for the tool parameter that is being retrieved. This must match the value passed into the RegisterToolParam method when the parameter was registered.

The value associated with the tool parameter is returned when calling this method. If the specified parameter does not exist or was not provided by the client, then an empty string is returned.

ProcessRequest Method (MCPServer Component)

Processes the request and sends the result.

Syntax

procedure ProcessRequest();

Remarks

When the Transport property is set to ttHTTP and the ProcessingMode property is set to modeExternalServer or modeOffline, this method acts as the main entry point for the component and is used to process a specific request. This method should not be called when the component is running in the modeEmbeddedServer mode.

Note that when using the modeOffline mode, the component processes the request and its associated headers supplied via the Request and RequestHeaders properties, respectively, and therefore ignores the parameters of this method.

Note that authentication occurs fully outside the scope of the component, and the logic that verifies these credentials is separate from any interaction with the component. Only once the request has been authenticated through this external verification should the be passed to this method.

ProcessRequests Method (MCPServer Component)

Instructs the component to start processing incoming requests.

Syntax

procedure ProcessRequests();

Remarks

This method instructs the component to process requests.

Notice that StartListening needs to be called before calling this method.

RegisterPrompt Method (MCPServer Component)

Registers a prompt that can be requested by the client.

Syntax

procedure RegisterPrompt(Name: String; Description: String);

Remarks

This method is used to register a new prompt that can be requested by the client.

Name specifies the unique name identifier for the prompt.

Description is critical in helping the client to understand the purpose and functionality of the prompt and should be set to a brief, human-readable description of what the prompt does.

Any prompt arguments associated with the prompt to be registered must be registered via the RegisterPromptArg method before this method is called.

Once the prompt has been registered, it is added to the Prompts properties and may be requested by the client, causing the PromptRequest event to be fired. Additionally, the RegisteredPromptArgs properties is cleared, and all prompt arguments previously in the properties are applied to the prompt.

Example: // Register two prompt argument schemas. server.RegisterPromptArg("code", "Code to explain", true); server.RegisterPromptArg("language", "Programming language", false); // Register the prompt with the added arguments. server.RegisterPrompt("explain-code", "Explain how code works");

Example: // Define argument schema. server.RegisterPromptArg("code", "Code to explain", true); server.RegisterPromptArg("language", "Programming language", false); // Register the prompt. string name = "explain-code"; string description = "Explain how code works"; server.RegisterPrompt(name, description); // Handling the prompt request. server.OnPromptRequest += (s, e) => { // Retrieve runtime values sent by client. string code = server.GetPromptParamValue("code"); string language = server.GetPromptParamValue("language") ?? "Unknown"; // Build response messages. // Assuming the client sent 'a = 1 + 2;' for the 'code' argument, and 'python' for the 'language' argument. // The following response should contain a user message with the text 'Explain how this python code works: \n\na = 1 + 2;'. server.AddPromptMessage("user", $"Explain how this {language} code works:\n\n{code}"); };

RegisterPromptArg Method (MCPServer Component)

Registers an argument for a prompt.

Syntax

procedure RegisterPromptArg(Name: String; Description: String; Required: Boolean);

Remarks

This method is used to register a new argument for a prompt that has not yet been registered via the RegisterPrompt method.

Name specifies the unique name identifier for the prompt argument.

Description identifies what information should be provided to this argument.

Required specifies whether the prompt argument must be supplied by the client when requesting the prompt.

Once the prompt argument has been registered, it is added to the RegisteredPromptArgs properties. All of the prompt arguments in this properties will be applied to the next prompt registered via RegisterPrompt, after which the properties is cleared.

Example: // Register two prompt argument schemas. server.RegisterPromptArg("code", "Code to explain", true); server.RegisterPromptArg("language", "Programming language", false); // Register the prompt with the added arguments. server.RegisterPrompt("explain-code", "Explain how code works");

RegisterResource Method (MCPServer Component)

Registers a resource that can be requested by the client.

Syntax

procedure RegisterResource(Uri: String; Name: String; Description: String);

Remarks

This method is used to register a new resource that can be requested by the client.

Uri specifies the unique identifier for the resource, of which common formats include file:///filename.ext for files, standard HTTP/HTTPS URLs for web resources, or custom schemes like db://table_name.

Name indicates the display name of the resource shown to end users by the client.

Description should be set to a brief, human-readable description of the resource's purpose, its content type, and appropriate use cases.

Once the resource has been registered, it is added to the Resources properties and may be requested by the client, causing the ResourceRequest event to be fired.

Note that once registered, resources become available for discovery, but the actual content (including format details and MIME type) is not loaded until a specific resource has been requested.

Resource URIs

Resource URIs represent a unique identifier for a resource and must follow the format: [scheme]://[host]/[path]

Common schemes include:

Custom URI Schemes

In addition to the previously listed schemes, applications may define their own custom URI schemes. These must be in accordance with RFC3986, especially with regards to reserved characters and general formatting recommendations. Characters such as :, /, and ? must be percent-encoded when used as data.

Client behavior may vary; therefore when registering resource URIs, servers must be prepared to handle every resource they register, regardless of client behavior. MCP-compatible clients may ignore scheme recommendations or fail to fetch resources as expected.

RegisterTool Method (MCPServer Component)

Registers a tool that can be requested by the client.

Syntax

procedure RegisterTool(Name: String; Description: String);

Remarks

This method is used to register a new tool that can be requested by the client.

Name specifies the unique name identifier for the tool.

Description is critical in helping the client to understand when to invoke this tool should be set to a brief, human-readable description of what the tool does.

Any tool parameters associated with the tool to be registered must be registered via the RegisterToolParam method before this method is called.

Once the tool has been registered, it is added to the Tools properties and may be requested by the client, causing the ToolRequest event to be fired. Additionally, the RegisteredToolParams properties is cleared, and all tool parameters previously in the properties are applied to the tool.

RegisterToolParam Method (MCPServer Component)

Registers a parameter for a tool.

Syntax

procedure RegisterToolParam(Name: String; Required: Boolean);

Remarks

This method is used to register a new parameter for a tool that has not yet been registered via the RegisterTool method.

Name specifies the unique name identifier for the tool parameter.

Required specifies whether the tool parameter must be supplied by the client when requesting the tool.

Once the tool parameter has been registered, it is added to the RegisteredToolParams properties. All of the tool parameters in this properties will be applied to the next tool registered via RegisterTool, after which the properties is cleared.

SendSamplingRequest Method (MCPServer Component)

Requests language model reasoning from the client.

Syntax

function SendSamplingRequest(): String;

Remarks

When the Transport property is set to ttStdio, this method is used to send a sampling request to the client, which is often used in order to utilize language model reasoning.

When called, the set of messages in the SamplingMessages properties will be passed over to the client in order for them to be used as a prompt for the client's language model. This method will then return the output of the prompt as text.

Example: server.OnToolRequest += (s, e) => { if (e.Name.Contains("summarize-text")) { // Grab text to summarize from the corresponding tool parameter. string text = GetText(); // Custom logic. // Add a prompt asking the client's LLM to summarize the text we just received. SamplingMessage message = new SamplingMessage(); message.Role = TRoles.rtUser; message.Text = "Summarize the following text: " + text; server.SamplingMessages.Add(message); // Send the sampling messages over to the client and request a response. string result = server.SendSamplingRequest(); // Return some text. server.AddToolMessage((int)TToolMessageTypes.mtText, result); } };

System Prompts

When performing a sampling request, the server may specify a system prompt to influence the client's LLM behavior during sampling. A system prompt provides high-level guidance or context for how the language model should interpret and respond to the messages in SamplingMessages. It is especially useful for establishing tone, role, or task-specific instructions.

Example: server.SystemPrompt = "You are an assistant meant to summarize text only using a formal tone."; SamplingMessage message = new SamplingMessage(); message.Role = TRoles.rtUser; message.Text = "Summarize the following text: " + text; server.SamplingMessages.Add(message); string result = server.SendSamplingRequest();

StartListening Method (MCPServer Component)

Instructs the component to start listening for incoming connections.

Syntax

procedure StartListening();

Remarks

When the Transport property is set to ttStdio, or it is set to ttHTTP and the ProcessingMode property is set to modeEmbeddedServer, this method instructs the component to start listening for incoming connections on the port specified by LocalPort.

To stop listening for new connections, please refer to the StopListening method.

StopListening Method (MCPServer Component)

Instructs the component to stop listening for new connections.

Syntax

procedure StopListening();

Remarks

When the Transport property is set to ttStdio, or it is set to ttHTTP and the ProcessingMode property is set to modeEmbeddedServer, this method instructs the component to stop listening for new connections. After being called, any new connection attempts will be rejected. Calling this method does not disconnect existing connections.

UnregisterPrompt Method (MCPServer Component)

Unregisters an existing prompt.

Syntax

procedure UnregisterPrompt(Name: String);

Remarks

This method is used to unregister a prompt that had previously been registered via the RegisterPrompt method. Once called, the specified prompt will be removed from the Prompts properties and will not be able to be requested by the client until it is registered again with RegisterPrompt.

Name specifies the unique name identifier for the prompt that will be unregistered. This should match the value passed into RegisterPrompt when the prompt was registered.

Example: // Register a prompt, making it available for clients to request. server.RegisterPromptArg("code-to-review", "The source code to review", true); server.RegisterPrompt("review-code", "Review the specified code"); // Handle client requests. // Unregister the prompt and prevent clients from being able to request it. server.UnregisterPrompt("review-code");

UnregisterResource Method (MCPServer Component)

Unregisters an existing resource.

Syntax

procedure UnregisterResource(Uri: String);

Remarks

This method is used to unregister a resource that had previously been registered via the RegisterResource method. Once called, the specified resource will be removed from the Resources properties and will not be able to be requested by the client until it is registered again with RegisterResource.

Uri specifies the unique identifier for the resource that will be unregistered. This should match the value passed into RegisterResource when the resource was registered.

Example: // Register a resource, making it available for clients to request. server.RegisterResource("file:///kb/test.txt", "Sample resource file", "A sample resource file with text content"); // Handle client requests. // Unregister the resource and prevent clients from being able to request it. server.UnregisterResource("file:///kb/test.txt");

Resource URIs

Resource URIs represent a unique identifier for a resource and must follow the format: [scheme]://[host]/[path]

Common schemes include:

Custom URI Schemes

In addition to the previously listed schemes, applications may define their own custom URI schemes. These must be in accordance with RFC3986, especially with regards to reserved characters and general formatting recommendations. Characters such as :, /, and ? must be percent-encoded when used as data.

Client behavior may vary; therefore when registering resource URIs, servers must be prepared to handle every resource they register, regardless of client behavior. MCP-compatible clients may ignore scheme recommendations or fail to fetch resources as expected.

UnregisterTool Method (MCPServer Component)

Unregisters an existing tool.

Syntax

procedure UnregisterTool(Name: String);

Remarks

This method is used to unregister a tool that had previously been registered via the RegisterTool method. Once called, the specified tool will be removed from the Tools properties and will not be able to be invoked by the client until it is registered again with RegisterTool.

Name specifies the unique name identifier for the tool that will be unregistered. This should match the value passed into RegisterTool when the tool was registered.

Example: // Register a tool, making it available for clients to invoke. server.RegisterTool("random-num", "Generate a random number"); // Handle client requests. // Unregister the tool and prevent clients from being able to invoke it. server.UnregisterTool("random-num");

Error Event (MCPServer Component)

Fires when an error occurs during operation.

Syntax

type TErrorEvent = procedure (
  Sender: TObject;
  ConnectionId: Integer;
  ErrorCode: Integer;
  const Description: String
) of Object;

property OnError: TErrorEvent read FOnError write FOnError;

Remarks

This event is fired when an unhandled exception is caught by the component, providing information about the error.

ConnectionId identifies the connection associated with the error.

ErrorCode contains the numeric error code representing the specific error condition.

Description contains a textual description of the error that occurred.

Log Event (MCPServer Component)

Fires once for each log message.

Syntax

type TLogEvent = procedure (
  Sender: TObject;
  LogLevel: Integer;
  const Message: String;
  const LogType: String
) of Object;

property OnLog: TLogEvent read FOnLog write FOnLog;

Remarks

This event is fired once for each log message generated by the component. The verbosity is controlled by the LogLevel configuration.

LogLevel indicates the detail level of the message. Possible values are:

Message is the log message.

LogType identifies the type of log entry. Possible values are as follows:

• NFS

PromptRequest Event (MCPServer Component)

Fires when a prompt is requested by the client.

Syntax

type TPromptRequestEvent = procedure (
  Sender: TObject;
  const Name: String;
  const Description: String
) of Object;

property OnPromptRequest: TPromptRequestEvent read FOnPromptRequest write FOnPromptRequest;

Remarks

This event is fired when a prompt has been requested by the client and allows the server to run custom logic. The text-based output to be returned to the client should be passed to the AddPromptMessage method while handling this event. This method may be called multiple times during the same prompt request to build a complete conversation history with alternating roles.

The GetPromptParamValue method may be called from within this event to retrieve the values for the prompt arguments associated with the requested prompt.

Name contains the unique name associated with the prompt.

Description holds the brief, human-readable description of what the prompt does.

Example: // Define argument schema. server.RegisterPromptArg("code", "Code to explain", true); server.RegisterPromptArg("language", "Programming language", false); // Register the prompt. string name = "explain-code"; string description = "Explain how code works"; server.RegisterPrompt(name, description); // Handling the prompt request. server.OnPromptRequest += (s, e) => { // Retrieve runtime values sent by client. string code = server.GetPromptParamValue("code"); string language = server.GetPromptParamValue("language") ?? "Unknown"; // Build response messages. // Assuming the client sent 'a = 1 + 2;' for the 'code' argument, and 'python' for the 'language' argument. // The following response should contain a user message with the text 'Explain how this python code works: \n\na = 1 + 2;'. server.AddPromptMessage("user", $"Explain how this {language} code works:\n\n{code}"); };

ResourceRequest Event (MCPServer Component)

Fires when a resource is requested by the client.

Syntax

type TResourceRequestEvent = procedure (
  Sender: TObject;
  const Uri: String
) of Object;

property OnResourceRequest: TResourceRequestEvent read FOnResourceRequest write FOnResourceRequest;

Remarks

This event is fired when a resource has been requested by the client and allows the server to run custom logic. The resource content to be returned to the client should be passed to the AddResourceContent method while handling this event. This method may be called multiple times during the same resource request to provide multiple related resources or different versions of the same content.

Uri specifies the unique resource identifier for the requested resource.

Example: server.OnResourceRequest += (s, e) => { string uri = e.Uri.ToString(); if (uri == "file:///logs/app.log") { string logData = File.ReadAllText("app.log"); server.AddResourceContent(uri, logData, "text/plain"); } else { server.Response = "[Unknown resource]"; } };

Resource URIs

Resource URIs represent a unique identifier for a resource and must follow the format: [scheme]://[host]/[path]

Common schemes include:

Custom URI Schemes

In addition to the previously listed schemes, applications may define their own custom URI schemes. These must be in accordance with RFC3986, especially with regards to reserved characters and general formatting recommendations. Characters such as :, /, and ? must be percent-encoded when used as data.

Client behavior may vary; therefore when registering resource URIs, servers must be prepared to handle every resource they register, regardless of client behavior. MCP-compatible clients may ignore scheme recommendations or fail to fetch resources as expected.

SessionEnd Event (MCPServer Component)

Fires when the component ends a session.

Syntax

type TSessionEndEvent = procedure (
  Sender: TObject;
  SessionId: Int64
) of Object;

property OnSessionEnd: TSessionEndEvent read FOnSessionEnd write FOnSessionEnd;

Remarks

When the Transport property is set to ttHTTP, this event fires when the component ends a previously started session. The component doesn't expect any errors to occur or be reported by handlers of this event.

SessionId specifies the unique identifier for the current session and can be obtained from the initial SessionStart event when the session is first opened.

SessionStart Event (MCPServer Component)

Fires when the component starts a session.

Syntax

type TSessionStartEvent = procedure (
  Sender: TObject;
  SessionId: Int64;
  var ResultCode: Integer
) of Object;

property OnSessionStart: TSessionStartEvent read FOnSessionStart write FOnSessionStart;

Remarks

When the Transport property is set to ttHTTP, this event fires when a new session with a connected client begins. The lifetime of a session is a single HTTP request and response, meaning once the response has been sent, the session ends.

SessionId is the unique identifier of the session, assigned by the component. It is provided as a parameter in all other events so that the operations for a particular client session may be tracked.

ResultCode provides an opportunity to report the operation as failed. This parameter will always be 0 when the event is fired, indicating the operation was successful. If the event cannot be handled in a "successful" manner for some reason, this must be set to a non-zero value.

Reporting an error will abort the operation.

ToolRequest Event (MCPServer Component)

Fires when a tool is requested by the client.

Syntax

type TToolRequestEvent = procedure (
  Sender: TObject;
  ConnectionId: Integer;
  const Name: String;
  const Description: String;
  var IsError: Boolean
) of Object;

property OnToolRequest: TToolRequestEvent read FOnToolRequest write FOnToolRequest;

Remarks

This event is fired when a tool has been requested by the client and allows the server to run custom logic. If applicable, the text-based output to be returned to the client should be passed to the AddToolMessage method while handling this event. This method may be called multiple times, once for each data type, if the tool needs to pass multiple types of data back to the client.

The GetToolParamValue method may be called from within this event to retrieve the values for the tool parameters associated with the requested tool.

Name identifies the unique name associated with the tool.

Description holds the brief, human-readable description of what the tool does, and while not necessary, is often used here for sampling.

IsError may be set to True to indicate when an error condition is met and the tool cannot be handled successfully.

Example: server.OnToolRequest += (s, e) => { if (e.Name.Contains("read-docs")) { // Execute custom logic. string documentation = ReadDocumentation(); // Return some text. server.AddToolMessage((int)TToolMessageTypes.mtText, documentation); } };

Certificate Type

This is the digital certificate being used.

Remarks

This type describes the current digital certificate. The certificate may be a public or private key. The fields are used to identify or select certificates.

• EffectiveDate
• ExpirationDate
• ExtendedKeyUsage
• Fingerprint
• FingerprintSHA1
• FingerprintSHA256
• Issuer
• PrivateKey
• PrivateKeyAvailable
• PrivateKeyContainer
• PublicKey
• PublicKeyAlgorithm
• PublicKeyLength
• SerialNumber
• SignatureAlgorithm
• Store
• StorePassword
• StoreType
• SubjectAltNames
• ThumbprintMD5
• ThumbprintSHA1
• ThumbprintSHA256
• Usage
• UsageFlags
• Version
• Subject
• Encoded

Fields

Constructors

>

constructor Create();

>

constructor Create(valEncoded: TBytes);

>

constructor Create(valStoreType: TsdmCertStoreTypes; valStore: String; valStorePassword: String; valSubject: String);

>

constructor Create(valStoreType: TsdmCertStoreTypes; valStore: TBytes; valStorePassword: String; valSubject: String);

Prompt Type

A registered prompt.

Remarks

This type represents a registered prompt that can be requested by the client.

• Description
• Name

Fields

Constructors

constructor Create();

PromptArg Type

A prompt argument to be registered.

Remarks

This type represents a prompt argument that is yet to be registered.

• Description
• Name
• Required

Fields

Constructors

constructor Create();

Resource Type

A registered resource.

Remarks

This type represents a registered resource that can be requested by the client.

• Description
• Mimetype
• Name
• Size
• Uri

Fields

Constructors

constructor Create();

SamplingMessage Type

An individual message in a sampling request.

Remarks

This type represents an individual message within a sampling request received by the server.

• Role
• Text

Fields

Constructors

constructor Create();

ServerSettings Type

The embedded HTTP server settings used by the component.

Remarks

This type describes the current embedded HTTP server settings used by the component. The fields are used to configure the server.

• AllowedAuthMethods
• LocalHost
• LocalPort
• Timeout

Fields

Constructors

>

constructor Create();

Tool Type

A registered tool.

Remarks

This type represents a registered tool that can be requested by the client.

• Description
• Name

Fields

Constructors

constructor Create();

ToolParam Type

A tool parameter to be registered.

Remarks

This type represents a tool parameter that is yet to be registered.

• ParamType
• Value

Fields

Constructors

constructor Create();

Config Settings (MCPServer Component)

The component accepts one or more of the following configuration settings. Configuration settings are similar in functionality to properties, but they are rarely used. In order to avoid "polluting" the property namespace of the component, access to these internal properties is provided through the Config method.

MCPServer Config Settings

Base Config Settings

Trappable Errors (MCPServer Component)

MCPServer Errors