mcpField
M
mcp-server

mcp-server

github.com/rrezartprebrezaDevelopment32888 installsUpdated 1w ago
Repository

About mcp-server

A Spring Boot 3 skill for implementing MCP servers with Spring AI or the raw MCP Java SDK. It helps Java developers select compatible dependencies, register tools, configure transports, and address security and testing concerns.

Key features

  1. 01Spring AI MCP starter guidance
  2. 02Tool callback registration
  3. 03stdio and Streamable HTTP setup
  4. 04Remote endpoint security checks
  5. 05MCP client verification guidance

Use cases

  1. 01Building an MCP server in Spring Boot
  2. 02Registering Spring AI tool methods
  3. 03Configuring local stdio or remote transport

MCP support

Covers how to expose tools through a Spring-based MCP server, using stdio for local clients or Streamable HTTP for remote clients where supported. It also describes testing discovery and tool calls with an MCP client.

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/rrezartprebreza/spring-boot-skills --skill mcp-server

After installing, type/mcp-serverin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

agents/
openai.yaml
examples/
bad-order-tools.java
good-order-tools.java
OrderMcpTools.java
templates/
mcp.json

Details

From the project README

MCP Server - Spring Boot 3

Prefer the Spring AI MCP server starters in a Spring Boot application. Let the Spring AI BOM manage its MCP SDK dependency; do not force the standalone SDK version into a starter-based application.

Dependencies

<!-- Pick exactly one transport starter. -->
<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-starter-mcp-server</artifactId>
</dependency>
<!-- Remote MVC: spring-ai-starter-mcp-server-webmvc -->
<!-- Remote reactive: spring-ai-starter-mcp-server-webflux -->

For a raw SDK application without Spring AI, use the current MCP Java SDK 2.x release and follow its migration guide. Do not assume raw SDK examples match the version managed by Spring AI 1.x.

<dependency>
    <groupId>io.modelcontextprotocol.sdk</groupId>
    <artifactId>mcp</artifactId>
    <version>2.0.0</version>
</dependency>

Spring AI 1.x tool callback integration

Spring AI 1.0.x exposes @Tool methods only through an explicit MethodToolCallbackProvider — a bare @Component with @Tool methods registers nothing. Spring AI 1.1.x additionally supports native @McpTool/@McpToolParam with spring.ai.mcp.server.annotation-scanner.enabled=true, imported from org.springaicommunity.mcp.annotation (Spring AI 2.0 moved them to org.springframework.ai.mcp.annotation). Pick one registration path per tool.

@Configuration
class McpToolsConfiguration {
    @Bean
    ToolCallbackProvider orderTools(OrderMcpTools tools) {
        return MethodToolCallbackProvider.builder()
            .toolObjects(tools)
            .build();
    }
}

@Component
final class OrderMcpTools {
    private final OrderService orderService;

    OrderMcpTools(OrderService orderService) {
        this.orderService = orderService;
    }

    @Tool(description = "Get an order by UUID with line items and status history")
    OrderResponse getOrder(
            @ToolParam(description = "Order UUID") String orderId) {
        return OrderResponse.from(orderService.findById(UUID.fromString(orderId)));
    }
}
  • Return DTOs rather than serialized JPA entities.
  • Register each tool object once; duplicate providers create duplicate capabilities.
  • Keep tool names and descriptions stable because clients and models depend on them.
  • Bound result size, execution time, and downstream calls.
  • For raw SDK code, use builders such as CallToolResult.builder(); old constructors were removed.

Transport configuration

spring:
  main:
    banner-mode: "off"
  ai:
    mcp:
      server:
        name: order-service-mcp
        version: 1.0.0
        type: SYNC
        stdio: true

Use stdio when a local client launches the jar. Use the matching MVC or WebFlux starter and Streamable HTTP for a remote server when supported by the selected Spring AI 1.x release. Keep stdout clean in stdio mode because it carries JSON-RPC frames.

Security and verification

  • Authenticate and authorize remote MCP endpoints and every underlying domain operation.
  • Validate identifiers, enum values, pagination, and result limits.
  • Do not expose secrets, stack traces, persistence entities, or unrestricted administrative tools.
  • Test discovery, invalid arguments, authorization, timeouts, and shutdown with a real MCP client.
  • Review the selected Spring AI 1.x release notes before copying 2.0 annotation examples.

Examples

  • See examples/OrderMcpTools.java, examples/good-order-tools.java, and examples/bad-order-tools.java.

Official sources

Gotchas

  • Agent overrides the MCP SDK under Spring AI 1.x - let the Spring AI BOM manage it.
  • Agent copies Spring AI 2.0 org.springframework.ai.mcp.annotation imports into a 1.1 project - 1.1 uses org.springaicommunity.mcp.annotation, and 1.0 has no MCP annotations at all.
  • Agent expects @Tool methods on a plain @Component to appear as MCP tools in 1.x - register them through a ToolCallbackProvider bean.
  • Agent passes --spring.ai.mcp.server.transport=stdio - no such property; Boot ignores it silently. Use spring.ai.mcp.server.stdio=true.
  • Agent uses removed new CallToolResult(...) constructors in raw SDK code - use builders.
  • Agent logs to stdout in stdio mode - route logs to stderr or a file and disable the banner.
  • Agent registers the same tool object twice - keep one callback provider per capability.
  • Agent exposes entities or unbounded collections - return bounded DTO contracts.

Comments

Markdown supported