"author:mahirfatih" matching MCP servers:

Estimates the token size of a message list and either drops the oldest messages or replaces them with a summary so requests stay under a local model's context window. This prevents oMLX/Qwen servers from rejecting calls with "Prompt is too long" errors.