Creating a token-aware system to manage the context window for LLMs is a sophisticated approach to optimizing multi-agent workflows. It correctly identifies a key bottleneck in current AI-powered development systems.

This is a TypeScript-based MCP server that provides a tool for counting tokens in files and directories.

Installation

To get started, you first need to clone the repository and build the server from the source code.

Step 1: Clone and Build

1. Clone the repository to your local machine:

   git clone https://github.com/intro0siddiqui/token-counter-server.git

2. Navigate into the directory:

   cd token-counter-server

3. Install dependencies and build the project:

   npm install
   npm run build

After this step, the server is ready to be used. You can now configure your MCP client (like Claude Desktop) to use it.

Step 2: Configure Your Client

You have two options for configuring your client:

Option A: Use the Full Path (Simple)

You can point your client directly to the built server file. This is the simplest method and doesn't require any global system changes.

In your claude_desktop_config.json, use the full path to build/index.js inside the directory you just cloned.

• MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "token-counter-server": {
      "command": "/path/to/your/cloned/repo/token-counter-server/build/index.js"
    }
  }
}

Remember to replace /path/to/your/cloned/repo with the actual, absolute path on your system.

Option B: Make the Command Globally Available (Advanced)

If you prefer to use a simple token-counter-server command without typing the full path, you can install the server as a global command.

Important: You must be inside the token-counter-server directory (the one you cloned) before running this command.

1. Install globally from the local source:

   npm install -g .

   This command creates a symbolic link on your system that points to the server executable.

2. Configure your client with the simple command: Now you can use token-counter-server in your client configuration:

   {
     "mcpServers": {
       "token-counter-server": {
         "command": "token-counter-server"
       }
     }
   }

Tools Included

This server provides the following tool:

• count_tokens: Counts the number of tokens in a file or directory.
  • path (string, required): The path to the file or directory.
  • file_pattern (string, optional): A glob pattern to filter files (e.g., *.ts).

Prompts

Here are some example prompts you can use:

• Count tokens in a single file:

  "Can you count the tokens in src/index.ts?"

• Count tokens in a directory with a filter:

  "How many tokens are in the src directory, only counting .ts files?"

You can find these and other examples in the prompts/ directory.

Resources

This MCP includes sample files for testing and demonstration purposes.

• resources/sample.txt: A sample text file to test token counting.

Development

• Install dependencies: npm install
• Build: npm run build
• Watch for changes: npm run watch

Debugging

To debug the server, you can use the MCP Inspector:

npm run inspector

This will provide a browser-based interface to inspect the communication between the client and the server.