swift-context-protocol

swift-context-protocol

A distributed context protocol implementation for Swift.

swift-context-protocol is a Swift-based implementation of the Model Context Protocol (MCP) for AI contexts. It leverages Swift’s distributed actor model to enable type-safe, asynchronous remote invocation of tools, resources, and prompts.

Features

• Distributed Context Protocol
  Define a common interface (ContextProtocol) for operations like ping, initialize, complete, set logging level, resource reading, tool calling, and prompt execution.

• Tool, Resource, and Prompt Protocols
  Provide standardized protocols for:

  • Tool: Executes a specific operation using JSON schema validated inputs.
  • Resource: Represents a resource identified by a URI, supporting asynchronous read operations.
  • Prompt: Generates prompt templates or responses based on input parameters.

• JSON Schema Support
  Input and output formats can be validated and described using JSONSchema.

• Distributed Actors
  Leverages Swift’s new distributed actor model (with WebSocketActorSystem) to allow remote invocation of protocol methods.

• Capability Negotiation & Initialization
  Exchange client and server information along with supported capabilities during the initialization handshake.

Installation

Add the following dependency to your Package.swift:

dependencies: [
    .package(url: "https://github.com/1amageek/swift-context-protocol.git", from: "1.0.0")
]

Then add "swift-context-protocol" to your target dependencies.

Usage

Creating a Custom Tool

Implement the Tool protocol to create a custom tool. For example, here's a simple echo tool that returns the input string:

import Foundation
import JSONSchema
import ContextProtocol

/// A simple tool that echoes the provided input.
public struct EchoTool: Tool {
    
    public typealias Input = Envelop<String>
    public typealias Output = String
    
    public var name: String = "echo"
    public var description: String = "A tool that echoes the provided input."
    public var inputSchema: JSONSchema? = .string()
    
    public var guide: String? = """
    # Tool Name
    EchoTool
    
    ## Description
    This tool echoes back the input string provided to it.
    
    ## Parameters
    - `input`: The string to be echoed.
      - **Type**: `String`
      - **Description**: The input text that will be returned as output.
    
    ## Usage
    Provide a valid string as input.
    
    ## Example
    ```swift
    let input = "Hello, world!"
    // EchoTool returns: "Hello, world!"
    ```
    """
    
    public func run(_ input: Envelop<String>) async throws -> String {
        return input.data
    }
}

Registering Components and Starting the Server

Create a distributed actor (e.g., ContextActor) that implements the ContextProtocol and registers your tools, resources, and prompts.

import Foundation
import Distributed
import WebSocketActors
import ContextProtocol

@main
struct Boot {
    
    static func main() async throws {
        let server = ContextServer()
        server.setTools([ EchoTool() ])
        try await server.start()
    }
}

Connecting a Client

You can connect to the server using a client wrapper like ContextClient:

import Foundation
import Distributed
import ContextProtocol

@main
struct ClientBoot {
    static func main() async throws {
        let client = try await ContextClient.connectServer()
        let data = Envelop(data: "Hello, World!")
        let parameters = try JSONEncoder().encode(data)
        let toolResponse = try await client.session.callTool(
            name: "echo",
            parameters: parameters
        )
        print(toolResponse)
    }
}

During connection, the client performs an initialization handshake where it sends its information and capabilities and receives server details and instructions. A cool connection banner is printed on success.

API Overview

ContextProtocol

The distributed actor protocol defines methods for:

• initialize: Handshake with client info and capabilities.
• ping: Simple connectivity check.
• complete: Trigger completion logic.
• setLoggingLevel: Adjust logging level.
• getPrompt: Request a prompt.
• listPrompts: List available prompt templates.
• listResources: Retrieve a list of resources.
• readResource: Read resource content (returns a ResourceContentData enum).
• subscribeResource / unsubscribeResource: Manage resource subscriptions.
• callTool: Call a tool and receive its output as a ResourceContentData enum.
• listTools: Retrieve a list of tools.
• sendRootsListChanged: Notify changes in the root resource list.

Domain Types

• ServerInfo / ClientInfo: Exchange basic identity and versioning information.
• CapabilityConfig: Key-value settings indicating supported capabilities.
• ResourceContentData: An enum (e.g., .text(String) or .binary(Data)) to represent resource contents.
• ToolResponse / PromptResponse / ResourceResponse: DTOs used to list available tools, prompts, and resources.
• ListResponse<T>: A generic wrapper for lists.

Contributing

Contributions are welcome! Please open issues or pull requests on GitHub.

License

This project is licensed under the MIT License.