pierre-filliolaud/lab_mcp_ai_agent_springboot

Эта страница переведена автоматически. Технические детали смотрите в оригинале. Открыть оригинал

1 stars 31 forks Качество 57 Тренд 5MIT

In this lab, you will build a real AI agent not a chatbot. The agent will understand user intent and take real actions on GitHub using MCP.

Обзор

In this lab, you will build a real AI agent not a chatbot. The agent will understand user intent and take real actions on GitHub using MCP. An AI Agent is a system that: • understands user intent, • reasons about what needs to be done, • and takes actions on external systems through controlled tools. 1. understand a natural language request, 2. reason with , 3. calls , 4. automatically create a , 5. be , , and . “Create a task to add OpenTelemetry and export traces via OTLP” All steps in this lab must follow , exactly as in a real engineering team. - - - Each issue must be: - created before coding - implemented on a dedicated branch - linked to commits - closed via a Pull Request - One PR per STEP - PR description must reference the issue - Issue is closed Create the following labels in the repository:

README

Показан фрагмент оригинального README. Открыть оригинал.

🧪 LAB – Building an MCP AI Agent with Claude

Java · Spring Boot · Gradle · LangChain4j · MCP · Claude · Testing · CI/CD · Docker · GitHub

In this lab, you will build a real AI agent not a chatbot. The agent will understand user intent and take real actions on GitHub using MCP.

This project implements an AI Agent. An AI Agent is a system that: • understands user intent, • reasons about what needs to be done, • and takes actions on external systems through controlled tools.

🎯 Global Lab Objective

Build step by step an AI agent able to:

understand a natural language request,
reason with Claude (Anthropic),
calls standardized tools via MCP,
automatically create a GitHub issue,
be tested, containerized, and integrated into CI/CD.

🧠 End-to-end Use Case

“Create a task to add OpenTelemetry and export traces via OTLP”

⟶ a well-structured GitHub issue is created automatically.

🏗️ Target Architecture

User
  ↓  HTTP POST /api/run
AgentController (Spring Boot)
  ↓
AgentService
  ↓
LangChain4j Agent (BacklogAgent)
  ↓  tool call (@Tool)
GitHubMcpTools (Spring Boot)
  ↓
McpHttpClient
  ↓  JSON-RPC over HTTP
MCP HTTP Wrapper (Node.js)
  ↓  STDIO
GitHub MCP Server (official)
  ↓
GitHub API (Issues)

flowchart LR
    %% =========================
    %% User / Client
    %% =========================
    User["👤 User(curl / UI)"]

    %% =========================
    %% Spring Boot App (Student)
    %% =========================
    subgraph SB["🔨 Spring Boot Applicationlab_mcp_ai_agent_springboot"]
        AgentController["🔨 AgentControllerREST APIPOST /api/run"]

        AgentService["🔨 AgentServiceOrchestrates Agent"]

        BacklogAgent["🔨 BacklogAgent(LangChain4j AI Service)System + User Prompt"]

        LangChainConfig["🔨 LangChainConfigAnthropicChatModelTool Wiring"]

        GitHubMcpTools["🔨 GitHubMcpTools@Tool createIssue(...)"]

        McpHttpClient["🔨 McpHttpClientJSON-RPC over HTTP"]
    end

    %% =========================
    %% External AI
    %% =========================
    Claude["📦 Anthropic Claude API(chat + tool-use)"]

    %% =========================
    %% MCP Layer
    %% =========================
    subgraph MCP["📦 MCP Layer"]
        MCPWrapper["🔨 MCP HTTP Wrapper(Node.js / Express)POST /mcp"]
        GitHubMCP["📦 GitHub MCP Server(STDIO, official)"]
    end

    %% =========================
    %% GitHub
    %% =========================
    GitHub["📦 GitHub API(Issues, Repos)"]

    %% =========================
    %% Flows
    %% =========================
    User -->|HTTP POST /api/run| AgentController
    AgentController --> AgentService
    AgentService --> BacklogAgent

    BacklogAgent -->|chat + tool schema| Claude
    Claude -->|tool call request| BacklogAgent

Открыть полный README на GitHub