token-savior

Level ALocal

MCP server that gets Claude to 97.9% (188/192) on a real coding benchmark at -80% active tokens and -83% wall time, vs 78.3% plain. Structural code navigation + persistent memory engine. Works with every MCP client.

Overview

Token Savior is an MCP server for AI coding agents that combines structural code navigation, a persistent memory engine, and Bash output compaction and rewriting. The project reports that on a 96-task coding benchmark it raised Claude Code's score from 78.3% to 97.9% while cutting active tokens by about 80% and wall time by about 83%. It works with every MCP client.

⭐ 1.2kSearch & InfoPythonRepo ↗Updated 2026-10-15

Installation

MCP client config (paste into mcp.json)

{
  "mcpServers": {
    "token-savior-recall": {
      "command": "uvx",
      "args": [
        "token-savior-recall"
      ]
    }
  }
}

Agent install prompt

Copy and paste into Cursor / Claude / other agents to install this MCP for you

Please read this MCP repository and install it into my current environment:
https://github.com/Mibayy/token-savior

Known working client config (can write into mcp.json):
{
  "mcpServers": {
    "token-savior-recall": {
      "command": "uvx",
      "args": [
        "token-savior-recall"
      ]
    }
  }
}

Please: 1) check README for deps and env vars; 2) write config for my client; 3) if secrets are missing, tell me what to fill — do not invent values.

Related servers