FastMCP 🚀

A fast, declarative, and type-safe server framework for the Model Context Protocol (MCP), inspired by FastMCP(Python).

Build powerful, production-ready MCP servers in Dart with minimal boilerplate. fastmcp uses modern language features and code generation to let you focus on your application logic, not the protocol details.

• ⚡️ Blazing Fast Development: Define MCP Tools, Resources, and Prompts as simple Dart functions using intuitive annotations.
• 🔒 Type-Safe by Design: The code generator automatically creates argument parsing and validation, eliminating manual casting and runtime errors.
• 🤖 Automatic Schema Generation: Your function signatures are automatically converted into compliant JSON schemas for tools and prompts.
• ✨ Advanced Features, Simple API: Built-in support for long-running tasks, progress reporting, and cancellation via a clean McpContext object.
• 💎 Modern & Compliant: Implements the latest MCP 2025-06-18 specification with a clean, modular architecture.
• 🔌 Pluggable Transports: Start with a simple StdioTransport for local development or use the production-ready HttpTransport for networked applications.

📚 Table of Contents

• 🚀 Getting Started
  • 1. Installation
  • 2. Create Your First Tool
  • 3. Run the Code Generator
  • 4. Run Your Server
• ✨ Features
  • Defining Resources
  • Defining Prompts
  • Long-Running Tasks with McpContext
  • Declarative Caching
  • Custom Metadata
• 🤝 Contributing
• ✍️ Author
• 📜 License

🚀 Getting Started

1. Installation

You'll need fastmcp for the core framework and fastmcp_generator + build_runner for code generation.

# Add to your pubspec.yaml dependencies
dart pub add fastmcp

# Add to your pubspec.yaml dev_dependencies
dart pub add --dev build_runner
dart pub add --dev fastmcp_generator

2. Create Your First Tool

In your main.dart (or any library file), create a FastMCP instance and annotate a function.

import 'package:fastmcp/fastmcp.dart';

part 'main.fastmcp.g.dart';

final app = FastMCP(name: 'MyFirstServer');

/// A simple tool that greets a user.
/// The doc comment is automatically used as the tool's description.
@Tool()
String greet({
  /// This description will auto assigned to the parameter
  required String name,
}) {
  return 'Hello, $name!';
}

void main() async {
  await app.run(
    registerGenerated: registerGenerated,
    transport: StdioTransport(),
  );
}

3. Run the Code Generator

From your project's root directory, run build_runner.

dart run build_runner build

This will generate the main.fastmcp.g.dart file, containing all the necessary boilerplate to connect your greet function to the MCP engine.

4. Run Your Server

That's it! Run your application.

$fastmcp/example dart run lib/main.dart
[INFO] 03:11:24.875 [McpEngine]: Registering generated functions...
[INFO] 03:11:24.877 [McpEngine]: Registration complete.
[INFO] 03:11:24.878 [McpEngine]: Engine connected to transport.
[INFO] 03:11:24.878 [McpEngine]: FastMCP server "FastMCP Showcase Server" is running...
[INFO] 03:11:24.884 [HttpTransport]: 🚀 HTTP server listening on http://0.0.0.0:8080/mcp

Your MCP server is now live, and clients can call the greet tool.

✨ Features

Defining Resources

Expose data to clients using the @Resource annotation. fastmcp automatically wraps simple return types.

/// Provides the current server time.
@Resource(
  uri: 'server://time',
  mimeType: 'text/plain',
)
String getCurrentTime() => DateTime.now().toIso8601String();

Defining Prompts

Create dynamic prompt templates for LLMs. Your function must return a PromptResult.

@Prompt(name: 'code-review')
PromptResult createCodeReviewPrompt({required String language}) {
  return PromptResult(
    description: 'A system prompt for a $language code reviewer.',
    messages: [
      Message(
        role: 'system',
        content: TextContent(
          text: 'You are an expert code reviewer for the $language language...'
        ),
      ),
    ],
  );
}

Long-Running Tasks with McpContext

For tasks that take time, simply add McpContext as a parameter to your tool. fastmcp will inject it, giving you access to progress reporting and cancellation.

@Tool()
Future<String> longTask(McpContext context, {int steps = 10}) async {
  for (int i = 1; i <= steps; i++) {
    // Check if the client cancelled the operation
    if (context.cancellationToken.isCancelled) {
      return 'Task Cancelled!';
    }
    // Report progress back to the client
    context.onProgress(i / steps, 'Processing step $i of $steps...');
    await Future.delayed(const Duration(seconds: 1));
  }
  return 'Task Complete!';
}

Custom Metadata

Add custom _meta fields to your tools and resources for advanced client integration.

@Tool(
  name: 'admin-tool',
  meta: const {
    'permissions': ['admin'],
    'dangerLevel': 'high',
  },
)
void performAdminAction() { /* ... */ }

🤝 Contributing

Contributions are welcome! Please feel free to open an issue or submit a pull request

✍️ Author

This project is maintained by Yash Makan.

I am currently looking for new job opportunities and exciting projects to work on. If you are looking for a dedicated Flutter developer or have an exciting project in mind, please feel free to reach out

• Email: contact@yashmakan.com
• Website: yashmakan.com
• LinkedIn: linkedin.com/in/yashmakan
• GitHub: @yashmakan
• Cal.com: @yashmakan

📜 License

This project is licensed under the MIT License. See the LICENSE file for more details.