Code Context MCP Server

An MCP (Model Context Protocol) server for analyzing codebases with accurate token counting. When connected to an LLM through MCP, this tool provides accurate token counts and file analysis features.

Installation

# From npm (global)
npx @cogniteration/code-context /path/to/directory

# Using bunx (no installation needed)
bunx @cogniteration/code-context /path/to/directory

Usage

Start the MCP server, specifying which directories it's allowed to access:

bunx @cogniteration/code-context /path/to/directory [/additional/directories...]

Tokenizer Selection

By default, the Claude tokenizer is used. You can specify a different tokenizer model:

bunx @cogniteration/code-context --tokenizer=Xenova/gpt-4 /path/to/directory

Available tokenizers:

• Xenova/claude-tokenizer (default)
• Xenova/gpt-4 - GPT-4 / GPT-3.5-turbo
• Xenova/mistral-tokenizer-v3 - Mistral v3
• And more (use --tokenizer=invalid to see full list)

Development

# Clone the repository
git clone https://github.com/AdjectiveAllison/cogniteration.git
cd cogniteration/code-context

# Install dependencies
bun install

# Run locally
bun start

# Development with auto-reload
bun run dev

License

MIT

Contributing

Issues and pull requests welcome at github.com/AdjectiveAllison/cogniteration.