FastMCP

The fastest way to build MCP servers in Swift.

If you found this helpful, you can support more open source work!

---

try await FastMCP.builder()
    .name("My Server")
    .addTools([WeatherTool()])
    .run()

That's it. Three lines to a working MCP server.

Why FastMCP?

• Zero boilerplate - No manual JSON-RPC handling, no protocol implementation
• Type-safe - Swift-native with full Sendable support
• Declarative - Fluent builder API that reads like configuration
• Complete - Tools, Resources, Prompts, and Sampling out of the box
• Production-ready - Graceful shutdown, logging, lifecycle hooks included

Installation

dependencies: [
    .package(url: "https://github.com/mehmetbaykar/swift-fast-mcp", from: "1.0.2")
]

30-Second Example

import FastMCP

struct WeatherTool: MCPTool {
    let name = "get_weather"
    let description: String? = "Get weather for a location"

    @Schemable
    struct Parameters: Sendable {
        let location: String
    }

    func call(with args: Parameters) async throws(ToolError) -> Content {
        [ToolContentItem(text: "Weather in \(args.location): 22°C, Sunny")]
    }
}

@main
struct MyServer {
    static func main() async throws {
        try await FastMCP.builder()
            .name("Weather Server")
            .addTools([WeatherTool()])
            .run()
    }
}

Build it. Run it. Connect it to Claude Desktop. Done.

Full Feature Set

Tools

AI-callable functions with automatic schema generation:

struct MathTool: MCPTool {
    let name = "calculate"
    let description: String? = "Perform math operations"

    @Schemable
    struct Parameters: Sendable {
        let operation: Operation
        let a: Double
        let b: Double
    }

    @Schemable
    enum Operation: String, Sendable {
        case add, subtract, multiply, divide
    }

    func call(with args: Parameters) async throws(ToolError) -> Content {
        let result = switch args.operation {
            case .add: args.a + args.b
            case .subtract: args.a - args.b
            case .multiply: args.a * args.b
            case .divide: args.a / args.b
        }
        return [ToolContentItem(text: "Result: \(result)")]
    }
}

Resources

Expose data to AI models:

struct ConfigResource: MCPResource {
    let uri = "config://app/settings"
    let name = "App Settings"
    let description: String? = "Application configuration"
    let mimeType: String? = "application/json"

    var content: Content {
        """
        {"theme": "dark", "version": "1.0.0"}
        """
    }
}

Prompts

Reusable conversation templates:

struct GreetingPrompt: MCPPrompt {
    let name = "greeting"
    let description: String? = "A greeting template"
    let arguments: [Prompt.Argument]? = [
        .init(name: "name", description: "Person's name", required: true)
    ]

    func getMessages(arguments: [String: Value]?) async throws -> [Prompt.Message] {
        let name = arguments?["name"]?.stringValue ?? "friend"
        return [
            .user("You are helping \(name)."),
            .assistant("Hello \(name)! How can I help?")
        ]
    }
}

Builder API

var logger = Logger(label: "my-server")
logger.logLevel = .debug

try await FastMCP.builder()
    .name("My Server")
    .version("1.0.0")
    .addTools([WeatherTool(), MathTool()])
    .addResources([ConfigResource()])
    .addPrompts([GreetingPrompt()])
    .enableSampling()
    .transport(.stdio)
    .logger(logger)
    .shutdownSignals([.sigterm, .sigint])
    .onStart { print("Started") }
    .onShutdown { print("Stopped") }
    .run()

Transport Options

Transport	Use Case
.stdio	Claude Desktop, CLI tools
.inMemory	Unit testing
.custom(transport)	Your own implementation

Claude Desktop

Add to claude_desktop_config.json:

{
    "mcpServers": {
        "my-server": {
            "command": "/path/to/my-server"
        }
    }
}

Requirements

• macOS 14+
• Swift 6.2+

Dependencies

• swift-sdk - Official MCP Swift SDK
• swift-mcp-toolkit - Tool/Resource abstractions
• swift-service-lifecycle - Graceful shutdown
• swift-log - Logging

Claude Code Integration

FastMCP ships with a Claude Code skill and a subagent that let Claude Code scaffold and build MCP server projects for you.

Setup

Copy the skills/ and .claude/ directories into your project:

# Copy the skill (project scaffolding)
cp -r skills/ .claude/skills/

# Copy the agent (expert assistance)
cp -r .claude/agents/ .claude/agents/

Usage

Scaffold a new project with the skill:

/swift-fast-mcp MyServer tools,resources,prompts

Claude generates a complete project with Package.swift, typed tools/resources/prompts, tests, and Claude Desktop configuration.

Get expert help with the subagent:

Claude automatically delegates to the swift-mcp-expert agent when you ask about MCPTool, MCPResource, MCPPrompt implementations, the builder API, @Schemable types, or testing patterns. You can also invoke it explicitly:

Use the swift-mcp-expert to help me build a weather tool

License

MIT