Anet MCP Server

A Rust implementation of the Model Control Protocol (MCP) server that enables communication between clients and AI models via a standardized protocol.

This project provides a scalable and asynchronous framework for building AI services using Rust, Tokio, and NATS. It is designed for developers building AI agent systems, LLM-based tools, or custom JSON-RPC 2.0 service layers. The architecture supports real-time message passing, making it ideal for microservices, AI orchestration, and tool-based model interaction.

Features

• ✅ JSON-RPC 2.0 compatible API
• 🔄 NATS transport layer for message passing
• 🛠️ Extensible tool system
• 🧠 Support for prompts and resources
• ⚡ Asynchronous request handling with Tokio

Requirements

• Rust 1.70+
• NATS server running locally or accessible via network

Installation

Add the following to your Cargo.toml:

[dependencies]
anet_mcp_server = "0.1.0"

Getting Started

Running the Example Server

The repository includes a basic example server that demonstrates core functionality:

# Start a NATS server in another terminal or ensure one is already running
# Example:
nats-server

# Run the example server
cargo run --example basic_server

Testing the Server

You can test the server using the included test client:

cargo run --example test_client

This will send various requests to the server and print the responses.

Usage

Creating a Server

use anet_mcp_server::{
    ServerBuilder, ServerCapabilities, 
    transport::nats::NatsTransport,
};
use serde_json::json;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let transport = NatsTransport::new("nats://localhost:4222", "mcp.requests").await?;

    let server = ServerBuilder::new()
        .transport(transport)
        .name("my-mcp-server")
        .version("0.1.0")
        .capabilities(ServerCapabilities {
            tools: Some(json!({})),
            prompts: Some(json!({})),
            resources: Some(json!({})),
            notification_options: None,
            experimental_capabilities: None,
        })
        .build()?;

    server.run().await
}

Implementing a Custom Tool

use anet_mcp_server::{Content, Tool};
use async_trait::async_trait;
use serde_json::{json, Value};

struct MyTool;

#[async_trait]
impl Tool for MyTool {
    fn name(&self) -> String {
        "my_tool".to_string()
    }

    fn description(&self) -> String {
        "A custom tool".to_string()
    }

    fn input_schema(&self) -> Value {
        json!({
            "type": "object",
            "properties": {
                "input": { "type": "string" }
            }
        })
    }

    async fn call(&self, input: Option<Value>) -> anyhow::Result<Vec<Content>> {
        Ok(vec![Content::Text {
            text: "Tool response".to_string()
        }])
    }
}

API Reference

The server implements the following JSON-RPC methods:

• initialize – Initialize the connection and get server information
• listTools – Get a list of available tools
• callTool – Call a specific tool with arguments
• listResources – Get a list of available resources
• readResource – Read a specific resource
• listPrompts – Get a list of available prompts
• getPrompt – Get a specific prompt with arguments

Architecture

The server follows a modular design:

• server – Core server logic and request handling
• transport – Message transport layer (currently NATS)
• tools – Tool interfaces and implementations
• types – Common data structures

License

MIT License

Let me know if you want badges, contribution guidelines, or example JSON-RPC payloads added to the README as well.