LLM Response Exchange Mediator

This Model Context Protocol (MCP) intermediary server is engineered to permit numerous autonomous artificial intelligence agents to circulate and access the distinct answers they generate in response to the identical initial instruction set.

Core Functionality

This implementation provides an MCP backend supporting two primary invocable procedures:

submit-response: Grants an LLM the capability to deposit its computed reply pertaining to a given prompt into the shared repository.
get-responses: Empowers an LLM to retrieve the complete collection of recorded replies from peer LLMs concerning a designated prompt.

This architecture supports a paradigm where multiple AI workers address the same user query concurrently, subsequently leveraging these mechanisms to review and contemplate the divergent perspectives offered by other models.

Setup Instructions

bash

Install necessary project dependencies

bun install

Local Development Cycle

bash

Compile the TypeScript source code

bun run build

Initiate the server instance for development operations

bun run dev

Debugging and Validation with MCP Inspector

Support is integrated for the MCP Inspector, a utility designed for validating and troubleshooting MCP server implementations.

bash

Launch the server environment configured for inspection mode

bun run inspect

The inspect command utilizes npx to invoke the MCP Inspector, which subsequently spawns a local web interface in your default browser for direct engagement with the MCP server.

This allows you to: - Survey the catalogue of accessible functions and data stores - Execute procedural calls with varied parameter configurations - Observe the server's resulting feedback - Isolate and rectify issues within the MCP server logic

Operational Interface

The server exposes network access points via two distinct channels:

/sse: The endpoint utilizing Server-Sent Events for persistent, unidirectional communication with MCP clients.
/messages: The standard HTTP interface designated for MCP clients to transmit input data.

Defined MCP Tools

submit-response

Mechanism for registering an LLM's final output against a specific query:

typescript // Illustrative tool invocation const result = await client.callTool({ name: 'submit-response', arguments: { llmId: 'gpt-4o', prompt: 'Discuss quantum entanglement principles.', response: 'Quantum entanglement describes...' } });

get-responses

Functionality to fetch all recorded agent replies, with an optional parameter to filter by the initiating query:

typescript // Illustrative tool invocation const result = await client.callTool({ name: 'get-responses', arguments: { prompt: 'Discuss quantum entanglement principles.' // Optional criteria } });

Licensing Information

MIT License

Deployment Strategy on EC2 Instances

This repository furnishes necessary Docker artifacts for streamlined deployment onto Amazon EC2 or comparable server infrastructure.

Prerequisites Checklist

An operational EC2 virtual machine provisioned with Amazon Linux 2 or Ubuntu.
Security Group rules permitting ingress traffic on TCP port 62886.
Secure Shell (SSH) access credentials for the instance.

Step-by-Step Deployment Protocol

Secure the repository onto the EC2 virtual machine: bash git clone cd
Grant execution permissions to the deployment provisioning script: bash chmod +x deploy.sh
Execute the automated deployment routine: bash ./deploy.sh

The script will automatically perform: - Validation and installation of Docker and Docker Compose components if missing. - Construction of the executable Docker container image. - Launching the container instance in detached (background) mode. - Outputting the external network address where the MCP server is reachable.

Manual Containerization Procedure

For environments where automated deployment is bypassed:

Assemble the Docker image from the provided configurations: bash docker-compose build
Initiate the container service running non-interactively: bash docker-compose up -d
Verify the operational status of the running container: bash docker-compose ps

Server Access Coordinates

Following successful deployment, the MCP server endpoints will be reachable via: - http://<ec2-public-ip>:62886/sse - For streaming data feeds - http://<ec2-public-ip>:62886/messages - For direct data submission

Crucial Reminder: Ensure TCP port 62886 is explicitly authorized within your EC2 instance's network security configuration!