ModelContextProtocol.jl [](https://svilupp.github.io/ModelContextProtocol.jl/stable/) [](https://svilupp.github.io/ModelContextProtocol.jl/dev/) [](https://github.com/svilupp/ModelContextProtocol.jl/actions/workflows/CI.yml?query=branch%3Amain) [](https://codecov.io/gh/svilupp/ModelContextProtocol.jl) [](https://github.com/JuliaTesting/Aqua.jl)

A Julia implementation of the Model Context Protocol (MCP), providing an SDK for building MCP-compliant servers and clients that enable AI models to access external tools.

Features

• JSON-RPC 2.0 compliant message handling
• Built-in support for tool registration and dynamic dispatch
• Simple API for creating MCP clients and servers

Installation

The package is not yet registered, so you need to add it manually:

using Pkg
Pkg.add(url="https://github.com/svilupp/ModelContextProtocol.jl.git")

Quick Start

Using MCP Tools with Claude

To use MCP-enabled tools with Claude, you need to configure your Claude desktop application:

1. Add tool configurations to your Claude desktop config file:
   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json

For detailed installation and setup instructions, visit the MCP Quickstart Guide.

Creating a MCP Server

using ModelContextProtocol

# Create a server
server = Server("MyServer", "1.0.0")

# Register a tool
register_tool!(server, "hello", 
    params -> Dict("result" => "Hello, $(get(params, "name", "World"))!"),
    Dict("type" => "object", "properties" => Dict("name" => Dict("type" => "string")))
)

# Run the server
run_server(server)

Using the MCP Client

using ModelContextProtocol

# Create a client
client = Client()

# Connect to a server
connect!(client, io_connection) # Can be stdin, a socket, etc.
initialize!(client)

# List available tools
tools = list_tools(client)
println("Available tools: ", [tool["name"] for tool in tools])

# Call a tool
response = call_tool(client, "hello", Dict("name" => "Julia"))

Included Example Tools

The package includes several example MCP-compliant tools:

Time Server

Provides tools for timezone conversion and current time queries:

• get_current_time: Get the current time in a specific timezone
• convert_time: Convert time between different timezones

Fetch Server

URL content retrieval with configurable options:

• fetch: Fetch content from URLs with options to limit content length and convert HTML to markdown

Weather Server

Weather forecast information:

• weather: Get weather forecasts for locations using the National Weather Service API

Translate Server

Language translation services:

• translate_text: Translate text between languages
• detect_language: Detect the language of provided text
• get_language_info: Get detailed information about a specific language

Example Config

For the following tools to be available in your Claude.ai desktop app, your config file (eg, ~/Library/Application Support/Claude/claude_desktop_config.json) should look like this:

{
    "mcpServers": {
        "fetch-server": {
            "command": "julia",
            "args": [
                "<your-path>/ModelContextProtocol.jl/examples/fetch/run_server.jl"
            ]
        },
        "time-server": {
            "command": "julia",
            "args": [
                "<your-path>/ModelContextProtocol.jl/examples/time/run_server.jl"
            ]
        },
        "weather-server": {
            "command": "julia",
            "args": [
                "<your-path>/ModelContextProtocol.jl/examples/weather/run_server.jl"
            ]
        },
        "translate-server": {
            "command": "julia",
            "args": [
                "<your-path>/ModelContextProtocol.jl/examples/translate/run_server.jl"
            ]
        }
    }
}

Interactive Example Client

The package includes an interactive client example that demonstrates how to use the MCP tools:

# Navigate to the example client directory
cd("examples/client")

# Activate the project
using Pkg
Pkg.activate(".")
Pkg.instantiate()

# Run the example client with a server
using ExampleClient
ExampleClient.main("/path/to/server_script.jl")

The interactive client supports commands like:

• get time utc
• convert time America/New_York Europe/Paris 2023-01-01T12:00:00
• get weather for New York
• fetch https://example.com

Development / Inspector

If you want to develop new servers or debug any existing ones, use the Inspector.

Just launch it npx @modelcontextprotocol/inspector, open it in the browser and connect to your server to inspect it.

For local tools, you must provide them as an argument to the inspector, eg, npx @modelcontextprotocol/inspector julia my_server/run_server.jl

Protocol Overview

The Model Context Protocol (MCP) standardizes interactions between AI models and tools. This implementation follows the core architecture using JSON-RPC 2.0 for message exchange.

Core Components

1. Server: Registers and provides tools, prompts, and resources
2. Client: Discovers and calls tools exposed by MCP servers
3. Transport: Handles message passing between clients and servers

License

This project is licensed under the MIT License - see the LICENSE file for details.