Quarkus + MCP = Agentic

What is Quarkus + MCP = Agentic?

This project uses Quarkus and the Model Context Protocol (MCP) to build an agentic application that orchestrates multiple MCP servers (Brave Search, Google Maps, Slack, filesystem) via Quarkus + LangChain4j. It is intended for developers who want to experiment with agentic AI using Java and a modular tool ecosystem.

How to use Quarkus + MCP = Agentic?

Install dependencies (node, npm, and a container runtime like Podman or Docker for telemetry). Create a .env file with API keys for Brave, OpenAI, Google Maps, and Slack. Create a playground directory for the filesystem MCP server. Run the application in dev mode using ./mvnw compile quarkus:dev and access the chat UI at http://localhost:8080 or the Dev UI at http://localhost:8080/q/dev/ to issue prompts.

Key features of Quarkus + MCP = Agentic

• Integrates multiple MCP servers (Brave, Google Maps, Slack, filesystem) in one app.
• Uses LangChain4j for LLM orchestration and tool calling.
• Provides a chat UI and a Dev UI for live interaction.
• Built-in telemetry with Grafana (requires container environment).
• Supports agentic reasoning for multi-step tasks.
• Packaged as a standard Quarkus application (JAR or native executable).

Use cases of Quarkus + MCP = Agentic

• Plan a team lunch: search for restaurants, check dietary needs, send Slack invitations, and generate a calendar file.
• Search the web with Brave and combine results with location data from Google Maps.
• Read and write files on the local filesystem (e.g., create ICS calendar files).
• Use the Dev UI to debug and inspect the LLM’s reasoning and tool usage.

FAQ from Quarkus + MCP = Agentic

What API keys are required?

You need API keys for Brave Web Search, OpenAI, Google Maps, and a Slack Bot User OAuth Token plus Team ID. These are set in an .env file or application.properties.

What runtime dependencies are needed?

You need Node.js and npm (to start MCP services), and a container environment such as Podman or Docker if you want telemetry with Grafana.

How does the agent communicate with MCP servers?

The application uses Quarkus + LangChain4j’s MCP client to connect to multiple MCP servers (stdio-based). The LLM decides which tool to invoke based on the prompt.

Where does data (e.g., created files) get stored?

When using the filesystem MCP server, files are created in a local playground directory (or a custom path configured in application.properties).

Can I run this in production?

The README notes that API keys should be stored securely (e.g., vaults or Kubernetes Secrets) for production. The application can be packaged as a JAR or native executable using standard Quarkus commands.