Token Compressor

Overview

Token Compressor is a Model Context Protocol server that reduces prompt size by 40-60% using a local LLM and embedding validation, while preserving all conditional logic. It integrates directly into AI assistants as callable tools, letting models invoke compression on your behalf.

The server runs entirely on your machine using the Python package token-compressor-mcp, installed via uvx token-compressor-mcp. No authentication or external services are required—it works out of the box. Version v0.1.1 is available now under the MIT license, with source code at base76-research-lab/token-compressor on GitHub.

Token Compressor belongs in the AI & Agents category and works with Claude Desktop, Claude Code, Cursor, VS Code, and other MCP clients. If you maintain this project, claim your listing here to verify ownership and keep information up to date.

Installation

// claude_desktop_config.json
{
  "mcpServers": {
    "token_compressor": {
      "command": "uvx",
      "args": [
        "token-compressor-mcp"
      ]
    }
  }
}