WithSeismic MCP Server Template

A production-ready MCP server template for hosting your own AI tools and prompts. Deploy remotely or run locally - built for developers who want to expose their tools to AI models without the infrastructure headaches.

://WHO_IS_THIS_FOR?

This template is perfect for:

• AI Tool Developers: Build and host your own tools that AI models can interact with
• Teams & Organizations: Set up a shared MCP server for your AI applications
• Individual Developers: Quick-start your MCP development with a production-ready setup
• API Providers: Expose your services to AI models through a standardized interface

Whether you want to:

• 🌐 Host a remote MCP server that multiple clients can connect to
• 🔧 Run locally via stdio for development and testing
• 🚀 Deploy a production-ready MCP implementation
• 🛠️ Build tools without worrying about infrastructure

This template provides everything you need to get started.

://OVERVIEW

The MCP server provides two ways to expose your tools to AI models:

1. Remote Server Mode (SSE): Deploy as a remote server that multiple clients can connect to
2. Perfect for production deployments (pnpm start:sse)
3. Supports multiple simultaneous connections
4. Easy deployment to platforms like Railway
5. Secure and scalable
6. Local Mode (stdio): Run locally for development and testing (pnpm start)
7. Great for local development
8. Direct stdio communication
9. Fast iteration and debugging

Key Features:

• Tools: Executable functions that models can call
• Prompts: Template-based message generators
• Auto-Discovery: Automatic registration of your components
• Type Safety: Full TypeScript support throughout

://WHY_USE_THIS_MCP_SERVER_TEMPLATE?

While the Model Context Protocol (MCP) is in its early stages, one thing is clear: rapid adoption requires frictionless development. This implementation is built with a singular focus: letting developers focus on building great tools, not infrastructure.

INSTANT_DEPLOYMENT

# Option 1: One-Click Deploy
Click the "Deploy on Railway" button above ☝️

# Option 2: Local Setup
pnpm install && pnpm dev

ZERO_FRICTION_DEVELOPMENT

• No Boilerplate: Define your tool, import it, and you're done. We handle the rest.
• Auto-Registration: Tools and prompts are automatically discovered and registered.
• Type Safety: Full TypeScript support with zero configuration.

BATTLE_TESTED_ARCHITECTURE

• Production Ready: Built for reliability and performance.
• Scalable Design: From simple tools to complex AI interactions.
• Error Resilient: Robust error handling and debugging built-in.

DEVELOPER_EXPERIENCE_FIRST

// That's it. This is all you need to create a tool.
export const myTool = new Tool(
  {
    name: "myTool",
    description: "What my tool does",
    inputSchema: z.object({ query: z.string() }),
    outputSchema: z.object({ result: z.string() }),
  },
  async (args) => {
    // Your logic here
    return { result: "Done!" };
  },
);

We handle:

• ⚡ Connection management
• 🔄 Tool registration
• 🛡️ Input validation
• 📦 State management
• 🚨 Error handling
• 📊 Logging and metrics

You focus on:

• 🎯 Building amazing tools
• 🧠 Creating powerful prompts
• 🚀 Shipping features

://GETTING_STARTED

INSTALLATION

pnpm install

RUNNING_THE_SERVER

Two modes are available:

1. Standard mode (stdio):

pnpm dev      # Development with hot reload
pnpm start    # Production

1. SSE (Server-Sent Events) mode:

pnpm dev:sse  # Development with hot reload
pnpm start:sse # Production

When running in SSE mode, connect to: http://localhost:3001/sse

://CORE_CONCEPTS

TOOLS

Tools are executable functions that models can invoke. Each tool:

• Has defined input/output schemas using Zod
• Is automatically registered with the registry
• Can perform any operation (calculation, file I/O, API calls, etc.)

Example tool:

import { z } from "zod";
import { Tool } from "../core";

const MyToolInputSchema = z.object({
  param1: z.string().describe("Parameter description"),
});

const MyToolOutputSchema = z.object({
  result: z.string().describe("Result description"),
});

export const myTool = new Tool(
  {
    name: "myTool",
    description: "What my tool does",
    inputSchema: MyToolInputSchema,
    outputSchema: MyToolOutputSchema,
  },
  async (args) => {
    const input = MyToolInputSchema.parse(args);
    // Tool logic here
    return { result: "processed result" };
  },
);

PROMPTS

Prompts are message generators that help structure model interactions. Each prompt:

• Defines its argument schema
• Generates messages in a consistent format
• Is automatically registered with the registry

Example prompt:

import { Prompt } from "../core";

export const myPrompt = new Prompt(
  {
    name: "myPrompt",
    description: "What my prompt does",
    arguments: [\
      {\
        name: "arg1",\
        description: "Argument description",\
        required: true,\
      },\
    ],
  },
  async (args) => {
    return [\
      {\
        role: "system",\
        content: {\
          type: "text",\
          text: `Generated message using ${args.arg1}`,\
        },\
      },\
    ];
  },
);

://ADDING_NEW_COMPONENTS

Creating a New Tool

1. Create a new file in src/modules/tools/
2. Define your input/output schemas using Zod
3. Create and export your tool instance
4. Add the export to src/modules/tools/index.ts

The registry will automatically:

• Register your tool
• Make it available to models
• Handle validation and error handling

Creating a New Prompt

1. Create a new file in src/modules/prompts/
2. Define your argument schema
3. Create and export your prompt instance
4. Add the export to src/modules/prompts/index.ts

The registry will automatically:

• Register your prompt
• Make it available to models
• Handle message generation and errors

Architecture

Core Components

• Registry: Central manager for all tools and prompts
• Tool: Base class for executable functions
• Prompt: Base class for message generators
• Server: MCP protocol implementation

Auto-Registration

The system uses a singleton Registry pattern that:

1. Automatically registers tools and prompts on import
2. Provides type-safe access to components
3. Handles all MCP protocol interactions

Error Handling

The system includes robust error handling:

• Type validation via Zod schemas
• Execution error wrapping
• Detailed error messages for debugging

Development

Type Safety

All components use TypeScript for full type safety:

• Input/output schemas are defined using Zod
• Type inference for tool arguments and results
• Comprehensive error types

Testing

Run tests using:

pnpm test

Best Practices

1. Always define clear input/output schemas
2. Use descriptive names and documentation
3. Handle errors gracefully
4. Follow the TypeScript guidelines in the codebase

://CONTRIBUTING

We welcome contributions! Please see our Contributing Guide for details on:

• Development workflow
• Code style guidelines
• Pull request process
• Issue reporting

Join our Discord community to connect with other contributors!

://SUPPORT

• 📫 Email: [email protected]
• 💬 Discord: Join our server

://LICENSE

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

© 2025 Doug, at WithSeismic dot com.