Agent Proxy for Event-Driven Agentic Choreography

Overview

This open-source project provides an Agent proxy that enables event-driven agentic choreography architecture. The project consists of two main modules:

1. Proxy Module – Implements an MCP-compliant and OpenAPI-compliant service that exposes Confluent Cloud topics.
2. Shell Module – Enables interaction with MCP services and Large Language Models (LLMs) such as OpenAI, Gemini, and AWS Bedrock.
3. Agentic Choreography Framework – A Java framework designed to accelerate the development of agents by taking care of infrastructure tasks like topic creation, schema registration, agent registration, and subscriptionRequest/subscriptionResponse handling. For more details, see the Agentic Choreography Framework.

Examples

1. Client Info Example - Demonstrates how an LLM can retrieve client information based on a user's first name.
2. JavaAgent Example – Demonstrates how a Java-based agent can integrate with the Agent Proxy using the Agentic Choreography Framework.
3. FlinkAgent Example – Demonstrates how to build an agent with Apache Flink that registers its capabilities with the proxy.

Key Features

• MCP Protocol Support: Allows LLM tool discovery and tool execution via stdio or HTTP SSE.
• Confluent/Kafka Integration: Provides seamless real-time data access via topics.
• OpenAPI Compliance: Standardized API definitions for easy integration.
• Multi-LLM Support: CLI module interacts with OpenAI, Gemini, and AWS Bedrock.

Architecture

How It Works

1. Agent Registration:

   • Each agent registers its capabilities with the Agent Proxy, allowing it to expose its services dynamically.
   • The proxy maintains a registry of available agent capabilities, making them discoverable for execution.

2. Service Invocation:

   • Once registered, agents can be invoked via the MCP Proxy using the MCP protocol.
   • Requests can be sent over stdio or HTTP SSE, enabling synchronous and asynchronous interactions.

3. Event-Driven Execution:

   • The Agent Proxy uses Kafka topics as an API layer to enable real-time data flow.
   • Agents can subscribe to relevant topics, process data, and publish results back into topics, ensuring a fully event-driven architecture

Diagram

   +----------------------+          +---------------------------+
   |       Agent 1        | <------> |        Agent Proxy        |
   +----------------------+          +---------------------------+
                                              |        ^
                                              v        |
   +----------------------+          +---------------------------+
   |       Agent 2        | <------> |          Topics           |
   +----------------------+          +---------------------------+

Key Benefits

• Dynamic Agent Discovery: Agents register their capabilities with the proxy, allowing them to be easily called when needed.
• Real-Time Event Processing: Topics enables scalable and reliable event-driven communication.
• Multi-Agent Collaboration: Multiple agents can work together, exchanging data and insights through the **Agent Proxy **.

This architecture provides a flexible and scalable foundation for agentic workflows by combining MCP, OpenAPI, and Topics.

Getting Started

Prerequisites

• Java 21+
• Maven
• Confluent Cloud account (with Kafka topics configured)

Installation

Clone the repository:

 git clone https://....
 cd mcp-openapi-proxy

Build the project:

mvn clean install

Setting Up the Environment

# follow the prompts
./set_env.sh

Running the Proxy

source .env # load environment variables set by set_env.sh
java -jar proxy/target/proxy-0.0.1-SNAPSHOT.jar

Using the SHELL

source .env # load environment variables set by set_env.sh
java -jar shell/target/shell-0.0.1-SNAPSHOT.jar

Configuration

Required Environment Variables

To run the proxy, set the following environment variables:

• BROKER_URL - The Confluent Cloud broker URL
• JAAS_USERNAME - The JAAS username for authentication
• JAAS_PASSWORD - The JAAS password for authentication
• SR_URL - Schema Registry URL
• SR_API_KEY - Schema Registry API key
• SR_API_SECRET - Schema Registry API secret

Optional Environment Variable

• REGISTRY_TOPIC - The topic for MCP registry (default: _agent_registry)
• CLIENT_ID - The unique client id. If not specified, a new UUID will be generated and saved into a .pid file in the current directory.

Contributing

Contributions are welcome! Feel free to open issues and submit pull requests.

License