A VS Code extension that displays all available MCP (Model Context Protocol) servers and calculates the token count of their tool descriptions using tiktoken.
Why Token Counting?
Token counts are more accurate than character counts for understanding the actual usage and cost of language model tool descriptions. This extension helps you:
Monitor the token usage of your MCP server configurations
Optimize tool descriptions to reduce token consumption
Understand the token overhead of different MCP servers
Features
MCP Server Overview: View all available MCP servers in your VS Code environment
Token Counting: Accurate token count calculation using tiktoken with cl100k_base encoding (same as GPT-4/GPT-3.5-turbo)
Tool Statistics: See the number of tools each server provides
Activity Bar Integration: Persistent view accessible from the activity bar
Real-time Refresh: Update counts dynamically as you add or remove MCP servers
Usage
Click the MCP Tools Token Counter icon in the activity bar (left sidebar)
The view will display all detected MCP servers with:
Server name
Number of tools
Total token count of all tool descriptions
Click the "Refresh" button to update the statistics after configuration changes
Use the search option to find a specific tool or server.
About Token Encoding
This extension uses the cl100k_base tokenizer, which is the encoding used by OpenAI's GPT-4 and GPT-3.5-turbo models. While this encoding is specifically designed for OpenAI models, it provides a reliable estimate for other language model providers as well:
Claude (Anthropic): Uses a similar BPE-based tokenizer with comparable token counts
Gemini (Google): Token counts typically within 10-20% of cl100k_base estimates
Other Models: Most modern LLMs use similar tokenization strategies
The cl100k_base encoding serves as an industry-standard reference point, giving you a practical estimate of token usage across different model providers, even if not perfectly exact.
Requirements
VS Code
MCP servers configured in your VS Code environment
Development
Setup
npm install
Compile
npm run compile
Watch Mode
npm run watch
Run Extension
Press F5 to open a new VS Code window with the extension loaded.
Technical Details
Uses vscode.lm.tools API to detect available language model tools
Implements webview view provider for persistent sidebar integration
Token counting powered by tiktoken with cl100k_base encoding
