Filters tools from upstream Multi-Modal Chat Protocol (MCP) servers using a regex-based deny list.
This proxy server addresses the critical issue of context pollution in large language models (LLMs) when interacting with Multi-Modal Chat Protocol (MCP) servers. Many LLMs currently fail to remove descriptions of denied or unused tools from their context, leading to significant token waste—potentially tens of thousands of tokens—and decreased efficiency. By acting as a transparent proxy, this tool effectively filters out unwanted or irrelevant tools based on regex-defined deny lists, ensuring that LLMs only receive necessary tool descriptions. This minimizes context engineering overhead and helps agents focus on relevant tasks without distraction from extraneous tool options.