TorstenS1/mcp-server

Spring AI MCP Server and Client

This project contains two main components:

• mcp-server: A Spring Boot application that implements a Model Context Protocol (MCP) server. It exposes tools to AI models and can convert OpenAPI specifications into MCP tools.
• mcp-client: A Spring Boot application that acts as a client to the MCP server. It dynamically discovers and uses tools exposed by the server to interact with a Large Language Model (LLM).
• python_mcp_server: A first hot Python implementation equivalent of Spring mcp-server.

Getting Started

This project is designed to be run using Docker Compose, which orchestrates all the services (Java MCP Server, Python MCP Server, Sample Tools App, and MCP Client).

Prerequisites

• Docker and Docker Compose: You must have Docker and Docker Compose installed on your system.
• OpenAI API Key: The mcp-client requires a valid OpenAI API key. Before running, edit mcp-client/config.yml and replace YOUR_OPENAI_API_KEY with your actual key.

Building Projects

To build all the Java projects and install Python dependencies, use the provided build script:

chmod +x build-all.sh
./build-all.sh

Running the Project

To start all services (Java MCP Server, Python MCP Server, Sample Tools App, and MCP Client) using Docker Compose:

docker-compose up --build

Once running, the services will be accessible on the following ports on your host machine:

• Java MCP Server: http://localhost:18001
• Python MCP Server: http://localhost:18002
• Sample Tools App: http://localhost:18003
• MCP Client: http://localhost:18004

How They Work Together

The mcp-server acts as a tool provider for the mcp-client. The client connects to the server to get a list of available tools, which it then uses to interact with the configured LLM. This allows for a separation of concerns, where the server manages the tools and the client focuses on the interaction with the LLM.

How to Register New Tools

Once you have a service with an OpenAPI specification, you can register it as a tool in either the Python or Java MCP server.

Python MCP Server

Add the tool to the tools list in the python_mcp_server/mcp_server.yml file. For example, to add the sample-tools-app:

mcp_server:
  host: 0.0.0.0
  port: 18002
  tools:
  - description: API for managing users and orders
    name: sample-tools
    rest_api_url: http://sample-tools-app:18003/v3/api-docs

Java MCP Server

The Java MCP Server can load tools on startup from an mcp_server.yml file located in its /config directory. An example is provided at mcp-server/config/mcp_server.yml.

Alternatively, you can dynamically register tools by sending a POST request to the /api/register-openapi endpoint. When running in Docker, use the service name (e.g., sample-tools-app) as the hostname for the source URL.

curl -X POST http://localhost:18001/api/register-openapi \
-H "Content-Type: application/json" \
-d '{
      "type": "URL",
      "source": "http://sample-tools-app:18003/v3/api-docs",
      "friendlyName": "sample-tools",
      "description": "Tools for managing users and orders"
    }'

End-to-End Testing

This project includes a full end-to-end test to verify that all components (mcp-server-java, sample-tools-app, and mcp-client) are working together correctly.

Prerequisites

1. Docker and Docker Compose: You must have Docker and Docker Compose installed.
2. OpenAI API Key: The mcp-client requires a valid OpenAI API key to function. Before running the test, you must edit the mcp-client/config.yml file and replace the placeholder YOUR_OPENAI_API_KEY with your actual key.

Running the Test

A shell script test-e2e.sh is provided to automate the entire process.

1. Make the script executable:

   chmod +x test-e2e.sh
   

2. Run the script:

   ./test-e2e.sh
   

The script will:

• Build and start all the necessary services in Docker.
• Wait for the services to initialize.
• Send a prompt to the mcp-client that requires it to use a tool from the sample-tools-app.
• Print the final response from the LLM.
• Shut down and clean up all the services.

Contributing

Contributions are welcome! Please feel free to submit a pull request.

License

This project is licensed under the MIT License. See the LICENSE file for more information.