Learning Adapter

Adaptive proxy server that intelligently optimizes MCP tool responses by learning which data fields are most valuable and filtering out noise to reduce token usage by up to 80%. Uses OpenAI's API to automatically analyze tool outputs and classify fields into essential identifiers, useful data, and technical metadata, then applies smart masking to return only relevant information while providing access to hidden fields on demand through an 'include' parameter. Designed to federate multiple MCP servers while dramatically reducing response sizes for token-sensitive workflows, with automatic context injection and persistent learning that improves filtering accuracy over time.

Source

Repository: https://github.com/sivachow/mcp-learning-adapter