| Filename | Latest commit message | Latest commit date |
|---|---|---|
- Implement MCP server with async/sync support and builder pattern configuration - Add Server stdio/SSE transport layers with robust error handling - Convert McpClient and McpServer to interfaces for better extensibility - Add sampling capabilities for LLM interactions (text/image) - Implement dynamic root management and runtime tool management - Add comprehensive prompt and resource capabilities - Add support for dynamic root management (add/remove roots at runtime) Architecture Changes - Introduce registration records for tools, resources, and prompts - Improve thread safety with concurrent collections - Enhance resource template handling - Implement builder patterns for Server/Client capabilities - Add support for server capabilities and implementation info Transport Layer: - Implement SseServerTransport with SSE endpoints - Add StdioServerTransport for process-based communication - Reorganize transport packages for better organization - Improve transport error handling and shutdown processes Documentation: - Add detailed MCP reference documentation - Include architecture diagrams and API documentation - Add comprehensive test coverage for all components - Include sample projects demonstrating stdio/SSE transports BREAKING CHANGES: Transport Package Reorganization - Move StdioClientTransport to o.s.ai.mcp.client.transport - Move SseClientTransport to o.s.ai.mcp.client.transport - Move ServerParameters to o.s.ai.mcp.client.transport - StdioClientTransport.errorHandler changed to private with setter method Migration: Use setErrorHandler() method instead of direct field access API Changes - Replace static factory methods with builder pattern - Update ServerCapabilities to use builder pattern - Make StdioClientTransport.errorHandler private with setter - Replace rootsListProvider with direct root management - Remove deprecated static factory methods Resolves #27,#25,#24,#11 Related to #12 |
||
| .. | ||
| src/main | ||
| pom.xml | ||
| README.md | ||
Spring AI MCP Sample Project
This sample project demonstrates the usage of the Spring AI Model Context Protocol (MCP) implementation. It showcases how to create and use MCP servers and clients with different transport modes and capabilities.
Overview
The sample provides:
- Two transport mode implementations: Stdio and SSE (Server-Sent Events)
- Server capabilities:
- Tools support with list changes notifications
- Resources support with list changes notifications (no subscribe support)
- Prompts support with list changes notifications
- Sample implementations:
- Two MCP tools: Weather and Calculator
- One MCP resource: System Information
- One MCP prompt: Greeting
Building the Project
./mvnw clean package
Running the Server
The server can be started in two transport modes, controlled by the transport.mode property:
Stdio Mode (Default)
java -Dtransport.mode=stdio -jar target/spring-ai-mcp-sample-0.3.0-SNAPSHOT.jar
The Stdio mode server is automatically started by the client - no explicit server startup is needed.
But you have to build the server jar first: ./mvnw clean install -DskipTests.
In Stdio mode the server must not emit any messages/logs to the console (e.g. standard out) but the JSON messages produced by the server.
SSE Mode
java -Dtransport.mode=sse -jar target/spring-ai-mcp-sample-0.3.0-SNAPSHOT.jar
Sample Clients
The project includes example clients for both transport modes:
Stdio Client (ClientStdio.java)
var stdioParams = ServerParameters.builder("java")
.args("-Dtransport.mode=stdio", "-jar",
"target/spring-ai-mcp-sample-0.3.0-SNAPSHOT.jar")
.build();
var transport = new StdioClientTransport(stdioParams);
var client = McpClient.using(transport).sync();
SSE Client (ClientSse.java)
var transport = new SseClientTransport(WebClient.builder().baseUrl("http://localhost:8080"));
var client = McpClient.using(transport).sync();
Available Tools
Weather Tool
- Name:
weather - Description: Weather forecast tool by location
- Parameters:
city: String - The city to get weather for
- Example:
CallToolResult response = client.callTool(
new CallToolRequest("weather", Map.of("city", "Sofia"))
);
Calculator Tool
- Name:
calculator - Description: Performs basic arithmetic operations
- Parameters (JSON Schema):
{ "operation": { "type": "string", "enum": ["add", "subtract", "multiply", "divide"], "description": "The arithmetic operation to perform" }, "a": { "type": "number", "description": "First operand" }, "b": { "type": "number", "description": "Second operand" } } - Example:
CallToolResult response = client.callTool(
new CallToolRequest("calculator",
Map.of("operation", "multiply", "a", 2.0, "b", 3.0))
);
Available Resources
System Information Resource
- URI:
system://info - Description: Provides basic system information including Java version, OS, etc.
- MIME Type: application/json
- Returns: JSON object containing: javaVersion, osName, osVersion, osArch, processors, timestamp
Available Prompts
Greeting Prompt
- Name:
greeting - Description: A friendly greeting prompt
- Parameters:
name: String (required) - The name to greet
- Returns: A personalized greeting message from an assistant
Client Usage Example
// Initialize client
client.initialize();
// Test connection
client.ping();
// List available tools
ListToolsResult tools = client.listTools();
System.out.println("Available tools: " + tools);
// Call weather tool
CallToolResult weather = client.callTool(
new CallToolRequest("weather", Map.of("city", "Sofia"))
);
System.out.println("Weather: " + weather);
// Call calculator tool
CallToolResult calc = client.callTool(
new CallToolRequest("calculator",
Map.of("operation", "multiply", "a", 2.0, "b", 3.0))
);
System.out.println("Calculation: " + calc);
// Access system info resource
ReadResourceResult sysInfo = client.readResource(
new ReadResourceRequest("system://info")
);
System.out.println("System Info: " + sysInfo);
// Use greeting prompt
GetPromptResult greeting = client.getPrompt(
new GetPromptRequest("greeting", Map.of("name", "John"))
);
System.out.println("Greeting: " + greeting);
// Close client
client.closeGracefully();
Server Usage Example
@Configuration
public class CustomMcpServerConfig {
@Bean
public McpAsyncServer mcpServer(McpTransport transport) {
// Configure server capabilities
var capabilities = McpSchema.ServerCapabilities.builder()
.resources(false, true) // Resource support with list changes notifications
.tools(true) // Tool support with list changes notifications
.prompts(true) // Prompt support with list changes notifications
.build();
// Create custom tool
var customTool = new ToolRegistration(
new McpSchema.Tool("custom-tool", "Description", Map.of("param", "String")),
(arguments) -> {
String param = (String) arguments.get("param");
return new CallToolResult(
List.of(new TextContent("Result: " + param)),
false
);
}
);
// Create custom resource
var customResource = new ResourceRegistration(
new McpSchema.Resource(
"custom://resource",
"Custom Resource",
"Description",
"application/json",
null
),
(request) -> new ReadResourceResult(
List.of(new TextResourceContents(
request.uri(),
"application/json",
"{\"data\": \"example\"}"
))
)
);
// Create custom prompt
var customPrompt = new PromptRegistration(
new McpSchema.Prompt(
"custom-prompt",
"Description",
List.of(new McpSchema.PromptArgument("input", "Description", true))
),
request -> {
String input = (String) request.arguments().get("input");
var message = new PromptMessage(
Role.ASSISTANT,
new TextContent("Response to: " + input)
);
return new GetPromptResult("Result", List.of(message));
}
);
// Create and configure the server
return McpServer.using(transport)
.info("Custom MCP Server", "1.0.0")
.capabilities(capabilities)
.tools(List.of(customTool))
.resources(Map.of("custom://resource", customResource))
.prompts(List.of(customPrompt))
.async();
}
}
Configuration
The application can be configured through application.properties:
transport.mode: Transport mode to use (stdio/sse)server.port: Server port for SSE mode (default: 8080)- Various logging configurations are available for debugging
Implementation Details
The sample demonstrates:
- Creating an MCP server with custom tools, resources, and prompts
- Configuring different transport modes
- Implementing tool handlers with JSON schema validation
- Resource implementations with dynamic content generation
- Prompt implementations with parameter handling
- Error handling and response formatting
- Synchronous client usage patterns