Show HN: A Proxy between LLMs and MCP servers with policy the model cannot reach (github.com)

A new project called extensible-mcp has been introduced as a proxy solution that interfaces between Large Language Models (LLMs) and Model Context Protocol (MCP) servers, emphasizing security and dynamic functionality. The key innovation is its ability to dynamically load servers during runtime, enabling LLMs to access a wider range of tools on demand without needing to restart or pre-configure tool definitions in prompts. By embedding these definitions into a vector index, rather than cluttering context windows, extensible-mcp mitigates issues like token bloat and enhances model performance. This architecture is significant for the AI/ML community as it shifts responsibility for security from the LLM to a deterministic filter pipeline. Such a design ensures that LLMs execute only those operations they have previously identified through a structured filtering mechanism, reducing the risk of prompt injection attacks. It also supports customizable security policies that can adapt to various use cases, promoting safe interactions between agents and external systems. With features like access control, Rego policy evaluation, and credential management, extensible-mcp addresses critical shortcomings in current LLM deployment while providing a robust foundation for future developments in secure AI-driven applications.