andreacomo/mcp-poc

Model Context Protocol POC

This project aims to play with MCP and Spring AI.

This is a Maven Multi Module project

MCP Server

STDIO based communication implementation. Based on Official walkthrough

The server exposes two tools:

• get_city_geolocation: get latitude and longitude from a city name
• get_weather_forecast: retrieve weather forecast for the given latitude and longitude. Default is 7 days

How to inspect the server

• build the project:

  mvn clean package
  

• Run the inspector (requires Node.js)

  npx @modelcontextprotocol/inspector
  

• Go to http://localhost:6274 and set:
  • Transport Type: STDIO
  • Command: java
  • Arguments: -jar /absolute/path/to/mcp-server-1.0.0-SNAPSHOT.jar
  • Connect

Now you can explore the MPC server

How to connect the MCP Server to Claude Desktop

• Locate the file claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
  • Linux not yet supported by Claude Desktop

• Add a new entry in the mcpServers object:

    {
      "mcpServers": {
        "spring-ai-mcp-weather": {
          "command": "java",
          "args": [
            "-jar",
            "/absolute/path/to/mcp-server-1.0.0-SNAPSHOT.jar"
          ]
        }
      }
   }
  

• Try to ask Claude "what's the forecast for tomorrow in [your city name]?". Claude will be capable of understand that need to retrieve geo info to give you the answer (using both tools)

  Please note that the prompt is bound to Italy cities

MCP Client

The client is basically a command line chat based on OpenAI: thanks to Spring AI and MCP integration, the client connects to the MCP servers defined in and create the clients automatically for each configured server.

Please note that the claude_desktop_config.json contains absolute paths because is using STDIO communication: this type of communication requires the client to start the server, so the project must me built first!

mvn package

The client requires the OpenAI key, provided as env variable OPENAI_API_KEY.

The client starts and list the available tools, then is ready to chat.