MCP Servers

A collection of Model Context Protocol (MCP) servers for personal use, built with TypeScript and the FastMCP framework.

Overview

This project provides MCP servers that can be used with AI assistants and other applications that support the Model Context Protocol. Currently includes tools for token counting and text analysis.

Features

Tiktoken Tool

• Token Counting: Determine the number of tokens in text using the js-tiktoken library
• Uses the o200k_base encoding (GPT-4 tokenizer)
• Useful for managing token limits in AI applications

Claude Desktop Integration

This package is designed to work with Claude Desktop's MCP Server configuration. You can use it directly with npx without needing to install it locally.

Quick Setup

Add the following to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "berrydev-mcp-servers": {
      "command": "npx",
      "args": [
        "-y",
        "@berrydev-ai/mcp-servers"
      ],
      "env": {}
    }
  }
}

After adding this configuration, restart Claude Desktop. The tiktoken tool will be available for token counting operations.

Installation

npm install

Publishing to NPM

This package is configured to be published to NPM for use with npx. The package includes:

• Executable binary configuration
• Automatic builds before publishing
• Proper file inclusion for distribution

npm publish

Development

Build the project

npm run build

Development mode (watch)

npm run dev

Run the server

npm start

Inspect with MCP Inspector

npm run inspector

Inspect with FastMCP CLI

npm run mcp-cli

Testing

Run tests:

npm test

Code Quality

Type checking

npm run type-check

Linting

npm run lint
npm run lint:fix  # Auto-fix issues

Formatting

npm run format        # Format code
npm run format:check  # Check formatting

Run all checks

npm run check

Project Structure

src/
├── index.ts          # Main server entry point
└── tools/
    ├── tiktoken.ts   # Token counting tool
    └── tiktoken.spec.ts # Tests for tiktoken tool

Available Tools

tiktoken

Counts tokens in the provided text using the GPT-4 tokenizer.

Parameters:

• text (string): The text to analyze (minimum 1 character)

Returns:

• Token count as a number

Example usage in Claude Desktop: Once configured, you can ask Claude to count tokens in text, and it will automatically use this tool.

Direct API usage:

{
  "name": "tiktoken",
  "arguments": {
    "text": "Hello, world!"
  }
}

CI/CD

The project includes GitHub Actions workflows for:

• Continuous Integration: Runs tests, linting, and type checking on Node.js 18.x and 20.x
• NPM Publishing: Automated publishing to NPM and GitHub Packages

Configuration

TypeScript

• Target: ES2022
• Module: NodeNext (for proper ES module support)
• Strict mode enabled
• Source maps and declarations generated

Dependencies

• Runtime: fastmcp, js-tiktoken, zod
• Development: TypeScript, ESLint, Prettier, Vitest

License

MIT

Author

Eric Berry (eric@berrydev.ai)