token compressor for Windsurf

io.github.base76-research-lab/token-compressor

Compress prompts 40-60% using local LLM + embedding validation. Preserves all conditionals.

client:Windsurf transport:stdio runtime:pypi

Install token compressor in Windsurf

~/.codeium/windsurf/mcp_config.json · windows: %USERPROFILE%\.codeium\windsurf\mcp_config.json

{
  "mcpServers": {
    "token-compressor": {
      "command": "uvx",
      "args": [
        "token-compressor-mcp"
      ]
    }
  }
}

Global only — Windsurf has no per-project MCP config. Refresh Cascade after saving. Windsurf docs