一个Java MCP（模型上下文协议）服务器，利用LSP4J包裹JDTLS（Eclipse JDT语言服务器），为像Claude这样的AI助手提供Java IDE功能。

特色

该MCP服务器提供以下工具：

先修条件

• Java 21+
• Maven 3.8+
• 安装JDTLS（例如，通过自制软件：brew install jdtls）

建造

这会在target/lsp4j-mcp-1.0.0-SNAPSHOT.jar处生成一个阴影JAR。

用途

选项1：运行脚本

选项2：直接Java命令

示例：

选项3：Claude代码MCP配置

补充到你的.mcp.json：

选项4：Docker

构建多阶段镜像。它包括Java 21、着色MCP服务器，以及 置顶Eclipse JDTLS分发：

在 stdio 上运行，并将 Java 项目挂载于 /workspace：

LSP4J-MCP Server

A Java MCP (Model Context Protocol) server that wraps JDTLS (Eclipse JDT Language Server) using LSP4J to provide Java IDE features to AI assistants like Claude.

Features

This MCP server exposes the following tools:

Prerequisites

• Java 21+
• Maven 3.8+
• JDTLS installed (e.g., via Homebrew: brew install jdtls)

Build

mvn clean package

This creates a shaded JAR at target/lsp4j-mcp-1.0.0-SNAPSHOT.jar.

Usage

Option 1: Run Script

./run.sh /path/to/your/java/project

Option 2: Direct Java Command

java -jar target/lsp4j-mcp-1.0.0-SNAPSHOT.jar  

Example:

java -jar target/lsp4j-mcp-1.0.0-SNAPSHOT.jar /Users/me/projects/myapp jdtls

Option 3: Claude Code MCP Configuration

Add to your .mcp.json:

{
  "mcpServers": {
    "java-lsp": {
      "command": "java",
      "args": [
        "-jar",
        "/path/to/LSP4J-MCP/target/lsp4j-mcp-1.0.0-SNAPSHOT.jar",
        "/path/to/your/java/project",
        "jdtls"
      ],
      "env": {
        "LOG_FILE": "/tmp/lsp4j-mcp.log"
      }
    }
  }
}

Option 4: Docker

Build the multi-stage image. It includes Java 21, the shaded MCP server, and a pinned Eclipse JDTLS distribution:

docker build -t lsp4j-mcp:local .

Run it over stdio and mount the Java project at /workspace:

docker run --rm -i \
  --mount type=bind,source=/absolute/path/to/java/project,target=/workspace \
  lsp4j-mcp:local

On Linux, use the host UID/GID when the language server needs to write project metadata into the mounted workspace:

docker run --rm -i \
  --user "$(id -u):$(id -g)" \
  --mount type=bind,source=/absolute/path/to/java/project,target=/workspace \
  lsp4j-mcp:local

Example MCP configuration:

{
  "mcpServers": {
    "java-lsp-docker": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "--mount",
        "type=bind,source=/absolute/path/to/java/project,target=/workspace",
        "lsp4j-mcp:local"
      ]
    }
  }
}

The server uses stdio, so the image exposes no network port. Container logs are written to stderr while stdout remains reserved for the MCP protocol. The image runs as UID/GID 10001 by default and expects at least 1 GB of container memory.

Project Structure

LSP4J-MCP/
├── pom.xml                              # Maven build configuration
├── run.sh                               # Startup script